fbpx

Featured This month

  • The Application of Machine Learning in Ophthalmology: The View from the Tech Side

    machine learning in ophthalmology
    Philip Marchenko
    30.11.2022
    15 min read

    According to the World Health Organization (WHO), artificial intelligence (AI) and machine learning (ML) will improve health outcomes by 2025. There are numerous digital technologies that shape the health of the future, yet AI and ML for medical image analysis look like one of the most promising innovations.

    The healthcare industry produces millions of medical images: MRI, CT, OCT, images from the lab, etc. The right diagnosis depends on the accuracy of the analysis by the specialists. Today AI can back up any medical specialist in medical image analysis: providing confidence and much-needed second opinion.

    Altris AI team decided to improve medical image analysis for just one type of medical image: Optical Coherence Tomography scans of the retina. To do it, the Altris AI team collected thousands of OCT scans and graphically labeled them, defining more than 100 pathologies and pathological signs. 

    machine learning in ophthalmology

    Then all this data was fed inside the AI model, and that is where the magic began… Further, I will tell how exactly we train the AI model of Altris AI so that it can detect more than 100 pathologies with 91% accuracy, but first, let’s discuss first why it is important for the healthcare industry.

    Why is automation with the help of AI important?

    Due to the delicate anatomy of the eye, its treatment carries a high risk of complications. Sometimes these complications can be the result of a medical error by an eye care specialist. But how often?

    According to the Altris team research, 20.2% of eye care practitioners miss minor, early, and rare pathologies on OCT scans 1- 3 times a week, and 4.4% miss them 3-5 times a week. But the worst thing is that 30.5% of ophthalmologists and optometrists are not even sure if they are missing any pathology at all.

    machine learning in ophthalmology

    Some medical errors may be minor, but some may cause significant harm to patients. Such medical errors can lead to medical malpractice lawsuits. That is why most ophthalmic clinics consider implementing AI to double-check the diagnosis of the ophthalmologist. 

    Besides, modern AI tools for medical image analysis have a high level of accuracy and can provide eye care specialists with a second opinion. 

    How to reach a high level of accuracy?

    It is almost always necessary to conduct many experiments to achieve a high level of model accuracy (in the case of Altris AI, it is 91%).  It is often done with the help of a machine learning pipeline.

    machine learning in ophthalmology

    How to reach a high level of accuracy:

    The machine learning pipeline is programmed by a team of engineers to perform certain steps automatically. It systematically trains and evaluates models, monitors experiments, and works with datasets.

      1. ML and Medical teams collect, annotate and preprocess data. It’s crucial to ensure the data quality is at its highest level because the model’s quality heavily depends on it. To do this, the teams developed a process and annotation guideline, which ensures that the number of errors in the annotation is minimized.
      2. ML team chooses the appropriate approach (model) depending on the collected data and the tasks. Each team member is well-versed in the most modern and high-quality approaches that solve emerging tasks.
      3. The selected model is trained on the annotated data.
      4. In the model evaluating and testing stage, we develop tests aimed at helping us understand whether the model is trained properly to perform the needed tasks.
      5. After the ML team is satisfied with the result, we deploy the model, which means the model is ready for production.
      6. While the model is running in production, we monitor its performance to ensure everything goes well.

    This workflow allows engineers to continuously fine-tune existing models alongside constant performance evaluations. The most significant advantage of this process is that it can be automated with the help of available tools. 

    What tasks does machine learning in ophthalmology have?

    Within the Altris AI platform, we solve 2 main tasks: segmentation and classification of OCT scans. 

    Classification task

    Classification is the task of determining which category a particular object belongs to. We assign each pathology to a certain class of pathologies (for example, glaucoma class).

    Segmentation task

    The image segmentation problem can be stated as the division of an image into regions that separate different objects from each other, and from the background.

    Key metrics of Altris ML pipeline

    When discussing classification and segmentation metrics in medical imaging machine learning, it is essential to mention the Confusion matrix (CM). CM is a visualization of our performance, which helps us understand whether the model is performing well in terms of predicted and real data. For a better explanation, let’s take a look at the picture. 

    machine learning in ophthalmology

    Let’s consider 4 possible outcomes from model predictions. Say we need to create a classifier to diagnose or predict if a patient has a disease (positive / 1 or TRUE) or not (negative/ 0 or FALSE). In such a case, the model can predict “yes” or “no”, and we can have an actual “yes” or “no”. Based on this, we can get 4 categories of results:

    • TP — true positive. The patient that actually has a disease has been diagnosed with this disease. A class was predicted to be true, and it is actually true.
    • TN — true negative. The patient is actually healthy and has been diagnosed as healthy. A class was predicted to be false, and it is actually false.
    • FP — false positive (type 1 error). The patient that is actually healthy has been diagnosed as having a disease. A class was predicted to be true, but it is actually false.
    • FN — false negative (type 2 error). The patient that actually has a disease has been diagnosed as healthy. A class was predicted to be false, but it is actually true.

    With the help of the confusion matrix, our ML engineers get specific metrics needed to train our medical imaging machine learning model properly. We discuss each metric in more detail in the following paragraphs.

    Classification metrics

    • Accuracy

    To find out how many of our predictions were correct, we divide the number of correct predictions by the total.

    machine learning in ophthalmology

    While being intuitive, the accuracy metric heavily relies on data specifics. If the dataset is imbalanced (the classes in a dataset are presented unevenly), we won’t get trustful results.

    For example, if we have a training dataset with 98% samples of class A (healthy patients) and only 2% samples of class B (sick patients). The medical imaging machine learning model can easily give you 98% training accuracy by predicting that every patient is healthy, even if they have a disease. Such results may have destructive consequences as people won’t get needed medical treatment.  

    • Precision

    Precision shows what proportion out of all positive predictions was correct.

    machine learning in ophthalmology

    Precision metric helps us in cases when we need to avoid False Negatives but can’t ignore False Positives. A typical example of this is a spam detector model. As engineers, we would be satisfied if the model sent a couple of spam letters to the inbox. However, sending an important non-spam letter to the spam folder (False Positive) is much worse.

    • Sensitivity/Recall

    Recall shows how many of all really sick patients we predicted and diagnosed correctly. It is a proportion of correctly positive predictions out of all positives.

    machine learning in ophthalmology

    In our case, you want to find all sick people, so it would not be so critical if the model diagnoses some healthy people as unhealthy. They would probably be sent to take some extra tests, which is annoying but not critical. But it’s much worse if the model diagnoses sick people as healthy and leaves them without treatment. 

      The sensitivity of Altris AI is 92,51%

    • Specificity

    The specificity shows how many of all healthy patients we predicted correctly. It is the proportion of actual negatives that the medical imaging machine learning model has correctly identified as such out of all negatives.

    machine learning in ophthalmology

    Specificity should be the metric of choice if you must cover all true negatives and you can’t tolerate any false positives as a result. For example, we’re making a fraud detector model in which all people whose credit card has been flagged as fraudulent (positive) will immediately go to jail. You don’t want to put innocent people behind bars, meaning false positives here are unacceptable. 

    The specificity of Altris AI is 99,80%

    Segmentation metrics

    Segmentation also can be thought of as a classification task. For each pixel, we make predictions about whether it is a certain object or not. Therefore, we can talk about Accuracy, Precision, Recall, and Specificity in terms of segmentation. 

    Let’s say we have a Ground Truth (what is really an object) and a Segmented (what the model predicted). The intersection in the picture below is the correct operation of the medical imaging machine learning model. All that is the difference (FN and FP) is the incorrect operation of the model. True negative (TN) is everything the model did not mark in this case.

    machine learning in ophthalmology

    Quite often, even after looking at such metrics, the problem of non-symmetricity remains in the segmentation tasks. For example, if we consider a tiny object, the Accuracy metric doesn’t work. Therefore, segmentation tasks also refer to additional metrics that allow taking into account the size of the object of the overall quality assessment. Let’s look at additional metrics in more detail.

    • Intersection over Union (IoU)/Jaccard

    Intersection over Union is an evaluation metric used to measure segmentation accuracy on a particular image. This metric is considered quite simple — the intersection zone is divided by the union of Ground Truth and Segmented.

    machine learning in ophthalmology

    Sometimes we get such results, like if the object was determined to be very large, but in fact, we see that it is small. Then the metric will be low, and vice versa. If the masks are approximately equal to each other, everything works correctly, and the metric will be high.

    • Dice score/F1

    The dice coefficient is 2 times the area of overlap divided by the total number of pixels in both images.

    machine learning in ophthalmology

    This metric is a slight modification of the previous one. The difference is that, in this case, we take the intersection area twice.

    Calculating scores over dataset

    We calculate the metrics described above for each scan. In order to count them over the entire dataset, we take each picture in this dataset, segment it, calculate the metric, and then take the average value of the metrics on each image.

    What is model validation in ML?

    In addition to evaluating the metrics, we also need to design the model validation procedure suitable for a specific task.

    When we have determined the metric that suits the task of machine learning for medical image analysis, we also need to understand what data to use for calculation. It will be wrong to calculate the metric on the training data because the model has already seen it. This means that we will not check the ability of the model to generalize in any way. Thus, we need a specific test dataset so that we can carry out quality control according to the selected metrics.

    The main tasks of the model validation are:

    • To provide an unbiased estimation of the accuracy that the model can achieve
    • To check whether the model is not overfitted

    Picking the correct model validation process is critical to guarantee the exactness of the validation method. In addition, there is no single suitable validation method for machine learning in ophthalmology — each task requires different validation. Engineers separately examine each task to see if data has leaked from the train dataset to the test dataset because this may lead to an overly optimistic estimate of the metrics.

    For example, we can take OCT images in different resolutions. We may need a higher image resolution for some diseases. If the medical imaging machine learning model overfits at the resolution of this OCT, it will be called a leak because the model should behave the same at any resolution.

    Overfitting and underfitting

    The model also has such an important property as a generalization. If the model did not see some data during training, it should not be difficult for the model to determine which class a certain image belongs to.

    At this stage, engineers may have two problems that they need to solve. The first problem is overfitting. When the model remembers the training data too well, we lose the ability to make correct predictions. The picture below illustrates this problem. The chart in the middle is a good fit when the model is general enough and has a positive trend, and the trend is well-learned. But the chart on the right shows a too-specific model that will not be able to guess the trend.

    machine learning in ophthalmology

    Another problem to solve is underfitting. This problem arises when we have chosen a model that is not complex enough to describe the trend in the data (left chart).

    Bias variance trade-off

    Another important concept we use in machine learning model validation is the bias variance trade-off. We want our models to always make accurate predictions and have no ground truth scattered. As shown at first/second circle.

    However, there are situations when we have a model that predicts something close to the target, but from dataset to dataset, it has a strong scatter. This is showcased in the second circle. 

    In circle three, you can observe a situation where the model has heap predictions on different datasets, but they are inaccurate. This situation usually indicates that we need to almost entirely rebuild the model.

    machine learning in ophthalmology

    Overfitting and bias variance trade-off are very important in working with the model, as they allow us to track errors and select a model that will balance between spread and hitting the target.

    Unbiased estimation

    In addition, within each model, we evaluate a set of parameters. We made a certain estimate (graph on the left), but in real life, the distribution of parameters differs (graph on the right). Thus, seeing that our estimate turned out to be shifted, we find another problem that needs to be solved. Machine learning in ophthalmology needs to make the estimate as unbiased as possible.

    machine learning in ophthalmology

    How do we validate the Altris AI model?

    There are three main steps in choosing a validation strategy for machine learning in ophthalmology:

    • we got familiar with ophthalmology, understood the nature of data, and where the leakages are possible;
    • We estimated the dataset size and target distribution;
    • understood the model’s training complexity (amount of operations/ number of parameters/ time) to pick the validation algorithms.

    After that, we have a reliable strategy for the machine learning model validation. Here are some fundamental concepts we use in the validation of models’ performance.

    Train/test split

    Train/test split is the most simple and basic strategy that we use to evaluate the model quality. This strategy splits the data into train and test and is used on small datasets. For example, we have a dataset of 1000 pictures, 700 of which we leave for training and 300 we take for the test.

    This method is good enough for prototyping. However, we don’t have enough datasets with it to do a simple double-check. This phenomenon is called high sampling bias: this happens when we encounter some kind of systematic error that did not fit into the distribution in the train or test.

    By dividing data into train and test, we are trying to simulate how the model works in the real world. But if we randomly split the data into train and test, our test sample will be far from the real one. This can be corrected by constructing several test samples from the number of data we have and examining the model performance. 

    Train/test/holdout set

    We leave the holdout as the final validation and use the train and test to work with the medical imaging machine learning model. After optimizing our model on the train/test split, we can check if we didn’t overfit it by validating our holdout set.

    machine learning in ophthalmology

    Using a holdout as a final sample helps us look at multiple test data distributions and see how much the models will differ.

    K-fold cross validation

    There is also a more general approach that Altris AI team use for validation — k-fold cross validation. This method divides all of our data equally into train and test.

    machine learning in ophthalmology

    We take the first part of the data and declare it as a test, then the second, and so on. Thus, we can train the model on each such division and see how it performs. We look at the variance and standard deviation of the resulting folds as it will give information about the stability of the model across different data inputs.

    Do we need ML models to perform on par with doctors?

    Here I will try to answer a question that worries many ophthalmologists and optometrists: can machine learning for medical image analysis surpasses an eye care specialist in assessing quality?

    In the diagram below, I have drawn an asymptote called the Best possible accuracy that can be achieved in solving a particular problem. We also have a Human level performance (HLP), which represents how a person can solve this problem. 

    HLP is the benchmark that the ML model strives for. Unlike the Best possible accuracy, for which there is no formula, HLP can be easily calculated. Therefore, we assume that if a model crosses the human quality level, we have already achieved the best possible quality for that model. Accordingly, we can try to approximate the Best possible accuracy with the HLP metric. And depending on this, we understand whether our model performs better or worse.

    machine learning in ophthalmology

    For those tasks that people do better and the ML model does worse, we do the following:

    • collect more data
    • run manual error analysis
    • do better bias/variance analysis

    But when the model crosses the HLP quality level, it is not entirely clear what to do next with the model and how to evaluate it further. So, in reality, we don’t need the model to outperform a human in interpreting images. We simply won’t know how to judge the quality of this model and whether it can be 100% objective and unbiased.

    Avoidable bias

    Let’s say we need to build a classifier for diabetic retinopathy based on OCT scans, and we have a control dataset prepared by people. In the first case, doctors are wrong 5% of the time. At the same time, the model on the train set is wrong in 10% of cases and on the test set — in 13%.

    machine learning in ophthalmology

    The difference between the model’s and the human’s performance is usually taken as the minimum difference between the train/test set and the human. In our case, it is 5% gap (10% – 5%) of avoidable bias. It is called avoidable bias because it can be fixed theoretically. In such a case, we need to take a more complex model and more data to better train the model.

    In the second case, doctors determine the disease with a 9% error. If the model defines a disease with the same rates as the previous example, then the difference between the train/test set and the human will be 1% (10% – 9%), which is much better than avoidable bias

    Looking at these two cases, we must choose a strategy that will lower the variance for the machine learning model so that it works stably on different test sets. Thus, taking into account the avoidable bias and the variance between the samples, we can build a strategy for training the model so that it could potentially outperform the HLP someday. However, do we need it now?

    Understanding HLP

    To better understand the HLP metric, let’s consider the task of determining dry AMD on OCT scans. We have a fixed dataset and 4 train sets, each one determining dry AMD with a specific accuracy:

    • ML engineers – 20%
    • ophthalmologists – 5%
    • 2 ophthalmologists – 3%
    • 2 ophthalmologists and 1 professor of ophthalmology – 2%

    machine learning in ophthalmology

    We take a result of 2% as the best HLP possible. To develop our model, we can choose the performances we strive to get. The 20% error result is irrelevant, so we discard this option. But the level of 1 doctor is enough for model version number 1 model. Thus, we are building a development strategy for model 1.

    Summing up

    Machine learning will revolutionize the eye care industry. It provides confidence and second opinion to eye care specialists in medical image analysis. 

    If you are looking for ways to use machine learning in your eye care practice, feel free to contact us. At Altris AI, we improve the diagnostic process for eye care practitioners by automating the detection of 54 pathological signs and 49 pathologies on OCT images.

  • Altris AI Builds Partnership with Academic Institutions

    Maria Znamenska
    21.11.2022
    2 min read

    At Altris AI we are determined to set higher diagnostic standards for the eye care industry, which is why we are searching for partnerships with academic institutions.

    We would be happy to provide full access to the Altris AI platform in the clinic and education settings with access to the severity differentiation, segmentation and classification, progression, and comparison analysis.

    We can also provide access to the additional feature which will be released over the coming months which includes retinal layer identification and separation, thickness, and volume measurement tools.

    What do you get?

    • The opportunity to test the unique artificial intelligence-powered platform for OCT scans analysis

    • Validation and feedback on the platform with an opportunity for our Head of Product and Chief Medical Officer to have calls with your team to discuss areas about the system development and performance

    • Work with Altris to gain specific scans of pathology that we request so we can train the model – obviously, always GDPR compliant.

    • The opportunity for us to use your logo on our website and promote the partnership with the University on our website, social media, etc.

    • Discuss additional ways that we can work together on future projects.

    Contact us

    Ask us any question
    Required field
    Required field

popular Posted

  • The Application of Machine Learning in Ophthalmology: The View from the Tech Side

    machine learning in ophthalmology
    Philip Marchenko
    30.11.2022
    15 min read

    According to the World Health Organization (WHO), artificial intelligence (AI) and machine learning (ML) will improve health outcomes by 2025. There are numerous digital technologies that shape the health of the future, yet AI and ML for medical image analysis look like one of the most promising innovations.

    The healthcare industry produces millions of medical images: MRI, CT, OCT, images from the lab, etc. The right diagnosis depends on the accuracy of the analysis by the specialists. Today AI can back up any medical specialist in medical image analysis: providing confidence and much-needed second opinion.

    Altris AI team decided to improve medical image analysis for just one type of medical image: Optical Coherence Tomography scans of the retina. To do it, the Altris AI team collected thousands of OCT scans and graphically labeled them, defining more than 100 pathologies and pathological signs. 

    machine learning in ophthalmology

    Then all this data was fed inside the AI model, and that is where the magic began… Further, I will tell how exactly we train the AI model of Altris AI so that it can detect more than 100 pathologies with 91% accuracy, but first, let’s discuss first why it is important for the healthcare industry.

    Why is automation with the help of AI important?

    Due to the delicate anatomy of the eye, its treatment carries a high risk of complications. Sometimes these complications can be the result of a medical error by an eye care specialist. But how often?

    According to the Altris team research, 20.2% of eye care practitioners miss minor, early, and rare pathologies on OCT scans 1- 3 times a week, and 4.4% miss them 3-5 times a week. But the worst thing is that 30.5% of ophthalmologists and optometrists are not even sure if they are missing any pathology at all.

    machine learning in ophthalmology

    Some medical errors may be minor, but some may cause significant harm to patients. Such medical errors can lead to medical malpractice lawsuits. That is why most ophthalmic clinics consider implementing AI to double-check the diagnosis of the ophthalmologist. 

    Besides, modern AI tools for medical image analysis have a high level of accuracy and can provide eye care specialists with a second opinion. 

    How to reach a high level of accuracy?

    It is almost always necessary to conduct many experiments to achieve a high level of model accuracy (in the case of Altris AI, it is 91%).  It is often done with the help of a machine learning pipeline.

    machine learning in ophthalmology

    How to reach a high level of accuracy:

    The machine learning pipeline is programmed by a team of engineers to perform certain steps automatically. It systematically trains and evaluates models, monitors experiments, and works with datasets.

      1. ML and Medical teams collect, annotate and preprocess data. It’s crucial to ensure the data quality is at its highest level because the model’s quality heavily depends on it. To do this, the teams developed a process and annotation guideline, which ensures that the number of errors in the annotation is minimized.
      2. ML team chooses the appropriate approach (model) depending on the collected data and the tasks. Each team member is well-versed in the most modern and high-quality approaches that solve emerging tasks.
      3. The selected model is trained on the annotated data.
      4. In the model evaluating and testing stage, we develop tests aimed at helping us understand whether the model is trained properly to perform the needed tasks.
      5. After the ML team is satisfied with the result, we deploy the model, which means the model is ready for production.
      6. While the model is running in production, we monitor its performance to ensure everything goes well.

    This workflow allows engineers to continuously fine-tune existing models alongside constant performance evaluations. The most significant advantage of this process is that it can be automated with the help of available tools. 

    What tasks does machine learning in ophthalmology have?

    Within the Altris AI platform, we solve 2 main tasks: segmentation and classification of OCT scans. 

    Classification task

    Classification is the task of determining which category a particular object belongs to. We assign each pathology to a certain class of pathologies (for example, glaucoma class).

    Segmentation task

    The image segmentation problem can be stated as the division of an image into regions that separate different objects from each other, and from the background.

    Key metrics of Altris ML pipeline

    When discussing classification and segmentation metrics in medical imaging machine learning, it is essential to mention the Confusion matrix (CM). CM is a visualization of our performance, which helps us understand whether the model is performing well in terms of predicted and real data. For a better explanation, let’s take a look at the picture. 

    machine learning in ophthalmology

    Let’s consider 4 possible outcomes from model predictions. Say we need to create a classifier to diagnose or predict if a patient has a disease (positive / 1 or TRUE) or not (negative/ 0 or FALSE). In such a case, the model can predict “yes” or “no”, and we can have an actual “yes” or “no”. Based on this, we can get 4 categories of results:

    • TP — true positive. The patient that actually has a disease has been diagnosed with this disease. A class was predicted to be true, and it is actually true.
    • TN — true negative. The patient is actually healthy and has been diagnosed as healthy. A class was predicted to be false, and it is actually false.
    • FP — false positive (type 1 error). The patient that is actually healthy has been diagnosed as having a disease. A class was predicted to be true, but it is actually false.
    • FN — false negative (type 2 error). The patient that actually has a disease has been diagnosed as healthy. A class was predicted to be false, but it is actually true.

    With the help of the confusion matrix, our ML engineers get specific metrics needed to train our medical imaging machine learning model properly. We discuss each metric in more detail in the following paragraphs.

    Classification metrics

    • Accuracy

    To find out how many of our predictions were correct, we divide the number of correct predictions by the total.

    machine learning in ophthalmology

    While being intuitive, the accuracy metric heavily relies on data specifics. If the dataset is imbalanced (the classes in a dataset are presented unevenly), we won’t get trustful results.

    For example, if we have a training dataset with 98% samples of class A (healthy patients) and only 2% samples of class B (sick patients). The medical imaging machine learning model can easily give you 98% training accuracy by predicting that every patient is healthy, even if they have a disease. Such results may have destructive consequences as people won’t get needed medical treatment.  

    • Precision

    Precision shows what proportion out of all positive predictions was correct.

    machine learning in ophthalmology

    Precision metric helps us in cases when we need to avoid False Negatives but can’t ignore False Positives. A typical example of this is a spam detector model. As engineers, we would be satisfied if the model sent a couple of spam letters to the inbox. However, sending an important non-spam letter to the spam folder (False Positive) is much worse.

    • Sensitivity/Recall

    Recall shows how many of all really sick patients we predicted and diagnosed correctly. It is a proportion of correctly positive predictions out of all positives.

    machine learning in ophthalmology

    In our case, you want to find all sick people, so it would not be so critical if the model diagnoses some healthy people as unhealthy. They would probably be sent to take some extra tests, which is annoying but not critical. But it’s much worse if the model diagnoses sick people as healthy and leaves them without treatment. 

      The sensitivity of Altris AI is 92,51%

    • Specificity

    The specificity shows how many of all healthy patients we predicted correctly. It is the proportion of actual negatives that the medical imaging machine learning model has correctly identified as such out of all negatives.

    machine learning in ophthalmology

    Specificity should be the metric of choice if you must cover all true negatives and you can’t tolerate any false positives as a result. For example, we’re making a fraud detector model in which all people whose credit card has been flagged as fraudulent (positive) will immediately go to jail. You don’t want to put innocent people behind bars, meaning false positives here are unacceptable. 

    The specificity of Altris AI is 99,80%

    Segmentation metrics

    Segmentation also can be thought of as a classification task. For each pixel, we make predictions about whether it is a certain object or not. Therefore, we can talk about Accuracy, Precision, Recall, and Specificity in terms of segmentation. 

    Let’s say we have a Ground Truth (what is really an object) and a Segmented (what the model predicted). The intersection in the picture below is the correct operation of the medical imaging machine learning model. All that is the difference (FN and FP) is the incorrect operation of the model. True negative (TN) is everything the model did not mark in this case.

    machine learning in ophthalmology

    Quite often, even after looking at such metrics, the problem of non-symmetricity remains in the segmentation tasks. For example, if we consider a tiny object, the Accuracy metric doesn’t work. Therefore, segmentation tasks also refer to additional metrics that allow taking into account the size of the object of the overall quality assessment. Let’s look at additional metrics in more detail.

    • Intersection over Union (IoU)/Jaccard

    Intersection over Union is an evaluation metric used to measure segmentation accuracy on a particular image. This metric is considered quite simple — the intersection zone is divided by the union of Ground Truth and Segmented.

    machine learning in ophthalmology

    Sometimes we get such results, like if the object was determined to be very large, but in fact, we see that it is small. Then the metric will be low, and vice versa. If the masks are approximately equal to each other, everything works correctly, and the metric will be high.

    • Dice score/F1

    The dice coefficient is 2 times the area of overlap divided by the total number of pixels in both images.

    machine learning in ophthalmology

    This metric is a slight modification of the previous one. The difference is that, in this case, we take the intersection area twice.

    Calculating scores over dataset

    We calculate the metrics described above for each scan. In order to count them over the entire dataset, we take each picture in this dataset, segment it, calculate the metric, and then take the average value of the metrics on each image.

    What is model validation in ML?

    In addition to evaluating the metrics, we also need to design the model validation procedure suitable for a specific task.

    When we have determined the metric that suits the task of machine learning for medical image analysis, we also need to understand what data to use for calculation. It will be wrong to calculate the metric on the training data because the model has already seen it. This means that we will not check the ability of the model to generalize in any way. Thus, we need a specific test dataset so that we can carry out quality control according to the selected metrics.

    The main tasks of the model validation are:

    • To provide an unbiased estimation of the accuracy that the model can achieve
    • To check whether the model is not overfitted

    Picking the correct model validation process is critical to guarantee the exactness of the validation method. In addition, there is no single suitable validation method for machine learning in ophthalmology — each task requires different validation. Engineers separately examine each task to see if data has leaked from the train dataset to the test dataset because this may lead to an overly optimistic estimate of the metrics.

    For example, we can take OCT images in different resolutions. We may need a higher image resolution for some diseases. If the medical imaging machine learning model overfits at the resolution of this OCT, it will be called a leak because the model should behave the same at any resolution.

    Overfitting and underfitting

    The model also has such an important property as a generalization. If the model did not see some data during training, it should not be difficult for the model to determine which class a certain image belongs to.

    At this stage, engineers may have two problems that they need to solve. The first problem is overfitting. When the model remembers the training data too well, we lose the ability to make correct predictions. The picture below illustrates this problem. The chart in the middle is a good fit when the model is general enough and has a positive trend, and the trend is well-learned. But the chart on the right shows a too-specific model that will not be able to guess the trend.

    machine learning in ophthalmology

    Another problem to solve is underfitting. This problem arises when we have chosen a model that is not complex enough to describe the trend in the data (left chart).

    Bias variance trade-off

    Another important concept we use in machine learning model validation is the bias variance trade-off. We want our models to always make accurate predictions and have no ground truth scattered. As shown at first/second circle.

    However, there are situations when we have a model that predicts something close to the target, but from dataset to dataset, it has a strong scatter. This is showcased in the second circle. 

    In circle three, you can observe a situation where the model has heap predictions on different datasets, but they are inaccurate. This situation usually indicates that we need to almost entirely rebuild the model.

    machine learning in ophthalmology

    Overfitting and bias variance trade-off are very important in working with the model, as they allow us to track errors and select a model that will balance between spread and hitting the target.

    Unbiased estimation

    In addition, within each model, we evaluate a set of parameters. We made a certain estimate (graph on the left), but in real life, the distribution of parameters differs (graph on the right). Thus, seeing that our estimate turned out to be shifted, we find another problem that needs to be solved. Machine learning in ophthalmology needs to make the estimate as unbiased as possible.

    machine learning in ophthalmology

    How do we validate the Altris AI model?

    There are three main steps in choosing a validation strategy for machine learning in ophthalmology:

    • we got familiar with ophthalmology, understood the nature of data, and where the leakages are possible;
    • We estimated the dataset size and target distribution;
    • understood the model’s training complexity (amount of operations/ number of parameters/ time) to pick the validation algorithms.

    After that, we have a reliable strategy for the machine learning model validation. Here are some fundamental concepts we use in the validation of models’ performance.

    Train/test split

    Train/test split is the most simple and basic strategy that we use to evaluate the model quality. This strategy splits the data into train and test and is used on small datasets. For example, we have a dataset of 1000 pictures, 700 of which we leave for training and 300 we take for the test.

    This method is good enough for prototyping. However, we don’t have enough datasets with it to do a simple double-check. This phenomenon is called high sampling bias: this happens when we encounter some kind of systematic error that did not fit into the distribution in the train or test.

    By dividing data into train and test, we are trying to simulate how the model works in the real world. But if we randomly split the data into train and test, our test sample will be far from the real one. This can be corrected by constructing several test samples from the number of data we have and examining the model performance. 

    Train/test/holdout set

    We leave the holdout as the final validation and use the train and test to work with the medical imaging machine learning model. After optimizing our model on the train/test split, we can check if we didn’t overfit it by validating our holdout set.

    machine learning in ophthalmology

    Using a holdout as a final sample helps us look at multiple test data distributions and see how much the models will differ.

    K-fold cross validation

    There is also a more general approach that Altris AI team use for validation — k-fold cross validation. This method divides all of our data equally into train and test.

    machine learning in ophthalmology

    We take the first part of the data and declare it as a test, then the second, and so on. Thus, we can train the model on each such division and see how it performs. We look at the variance and standard deviation of the resulting folds as it will give information about the stability of the model across different data inputs.

    Do we need ML models to perform on par with doctors?

    Here I will try to answer a question that worries many ophthalmologists and optometrists: can machine learning for medical image analysis surpasses an eye care specialist in assessing quality?

    In the diagram below, I have drawn an asymptote called the Best possible accuracy that can be achieved in solving a particular problem. We also have a Human level performance (HLP), which represents how a person can solve this problem. 

    HLP is the benchmark that the ML model strives for. Unlike the Best possible accuracy, for which there is no formula, HLP can be easily calculated. Therefore, we assume that if a model crosses the human quality level, we have already achieved the best possible quality for that model. Accordingly, we can try to approximate the Best possible accuracy with the HLP metric. And depending on this, we understand whether our model performs better or worse.

    machine learning in ophthalmology

    For those tasks that people do better and the ML model does worse, we do the following:

    • collect more data
    • run manual error analysis
    • do better bias/variance analysis

    But when the model crosses the HLP quality level, it is not entirely clear what to do next with the model and how to evaluate it further. So, in reality, we don’t need the model to outperform a human in interpreting images. We simply won’t know how to judge the quality of this model and whether it can be 100% objective and unbiased.

    Avoidable bias

    Let’s say we need to build a classifier for diabetic retinopathy based on OCT scans, and we have a control dataset prepared by people. In the first case, doctors are wrong 5% of the time. At the same time, the model on the train set is wrong in 10% of cases and on the test set — in 13%.

    machine learning in ophthalmology

    The difference between the model’s and the human’s performance is usually taken as the minimum difference between the train/test set and the human. In our case, it is 5% gap (10% – 5%) of avoidable bias. It is called avoidable bias because it can be fixed theoretically. In such a case, we need to take a more complex model and more data to better train the model.

    In the second case, doctors determine the disease with a 9% error. If the model defines a disease with the same rates as the previous example, then the difference between the train/test set and the human will be 1% (10% – 9%), which is much better than avoidable bias

    Looking at these two cases, we must choose a strategy that will lower the variance for the machine learning model so that it works stably on different test sets. Thus, taking into account the avoidable bias and the variance between the samples, we can build a strategy for training the model so that it could potentially outperform the HLP someday. However, do we need it now?

    Understanding HLP

    To better understand the HLP metric, let’s consider the task of determining dry AMD on OCT scans. We have a fixed dataset and 4 train sets, each one determining dry AMD with a specific accuracy:

    • ML engineers – 20%
    • ophthalmologists – 5%
    • 2 ophthalmologists – 3%
    • 2 ophthalmologists and 1 professor of ophthalmology – 2%

    machine learning in ophthalmology

    We take a result of 2% as the best HLP possible. To develop our model, we can choose the performances we strive to get. The 20% error result is irrelevant, so we discard this option. But the level of 1 doctor is enough for model version number 1 model. Thus, we are building a development strategy for model 1.

    Summing up

    Machine learning will revolutionize the eye care industry. It provides confidence and second opinion to eye care specialists in medical image analysis. 

    If you are looking for ways to use machine learning in your eye care practice, feel free to contact us. At Altris AI, we improve the diagnostic process for eye care practitioners by automating the detection of 54 pathological signs and 49 pathologies on OCT images.

  • Altris AI Builds Partnership with Academic Institutions

    Maria Znamenska
    21.11.2022
    2 min read

    At Altris AI we are determined to set higher diagnostic standards for the eye care industry, which is why we are searching for partnerships with academic institutions.

    We would be happy to provide full access to the Altris AI platform in the clinic and education settings with access to the severity differentiation, segmentation and classification, progression, and comparison analysis.

    We can also provide access to the additional feature which will be released over the coming months which includes retinal layer identification and separation, thickness, and volume measurement tools.

    What do you get?

    • The opportunity to test the unique artificial intelligence-powered platform for OCT scans analysis

    • Validation and feedback on the platform with an opportunity for our Head of Product and Chief Medical Officer to have calls with your team to discuss areas about the system development and performance

    • Work with Altris to gain specific scans of pathology that we request so we can train the model – obviously, always GDPR compliant.

    • The opportunity for us to use your logo on our website and promote the partnership with the University on our website, social media, etc.

    • Discuss additional ways that we can work together on future projects.

    Contact us

    Ask us any question
    Required field
    Required field
  • Will Artificial Intelligence Replace Ophthalmologists & Optometrists: Top 5 AI Misconceptions

    artificial intelligence replace ophthalmologists
    Maria Znamenska
    17.11.2022
    8 min read

    Back in 2019, at the RANZCO World Science Congress, Peter van Wijngaarden, Deputy Director of the Center for Eye Research, claimed that the eye sector is one of the leading areas of medicine in terms of artificial intelligence (AI) implementation. According to RANZCO, AI systems are already achieving incredible results and, in some cases, can even rival eye care specialists.

    There are a lot of similar examples when famous professors and specialists in the field of ophthalmology made predictions that AI is rapidly gaining strength in the eyecare industry. This gives rise to many myths and fears around the introduction of AI in clinical practice. More and more eye care professionals have faced the question: will artificial intelligence replace ophthalmologists and optometrists in the near future?

    The increased attention to the issue of optometrists and ophthalmologists replaced by AI was also provoked by a World Economic Forum (WEF) report. According to this report, people can lose 85 million jobs by 2025 due to the shifting division of labor between people and machines.

    In this post, we will discuss the top 5 misconceptions about AI that are most often faced by the owners of ophthalmological clinics and optometry centers in order to dispel them once and for all.

    Do AI algorithms work exactly like a human brain?

    artificial intelligence replace ophthalmologists

    The concept of optometrists and ophthalmologists replaced by robots is gaining popularity. Nowadays, eye care specialists often discuss the potential of AI training in human cognitive skills. It is no longer just about the ability of AI to detect Diabetic retinopathy or interpret OCT scans with greater accuracy. The question is, will AI ever be able to replicate human consciousness? And can AI replicate how the human brain works?

    What do we know about such models in different areas? AI systems are already demonstrating the work of some human cognitive functions. For example, AI models successfully compete with humans in computer games by gradually learning successful strategies. There is also an AI ​​model which creates enjoyable melodic music.

    However, optometrists and ophthalmologists replaced by AI still seem unrealistic. Even with the above examples mimicking some aspects of human behavior, an AI algorithm still needs to learn what empathy is. Artificial intelligence does not understand and cannot make sense of its surrounding, nor can it learn from its surrounding as humans do. The most famous example that confirms this inability of AI is Siri or Alexa. Voice assistants can set up appointments but give strange answers when the conversation goes differently than their scenario.

    While the human brain inspires modern AI techniques such as neural networks (NNs), the structure of NNs architectures is not biologically realistic. 

    First of all, there is a set of qualities that ophthalmologists and optometrists use every day. It is empathy for the patient, as well as creativity, teamwork, and adaptability. These qualities help doctors provide effective care to their patients. It is unlikely that the machines will ever be able to work with children, older adults, or patients with specific disabilities on par with humans. In addition, any patient would like to hear the diagnosis or discuss a treatment plan with a doctor, not a machine. 

    Therefore AI algorithm can’t work like a human brain, and the scenario where artificial intelligence replace ophthalmologists and optometrists will never happen. Nowadays, there are no developments that would make us think that AI image interpretation will ever be able at least to repeat important qualities of eye care specialists.

    Is today’s state of AI dangerous for humans?

    artificial intelligence replace ophthalmologists

    Today, AI algorithms can interpret retinal images and distinguish pathological from non-pathological scans. However, not all attempts at AI implementation have succeeded as well. One of the most popular non-medical examples is Facebook. Some time ago, Facebook tried to identify relevant news for certain groups of users. But the automated process could not detect the difference between real and fake news. Russian hackers managed to trick the system and bypass automatic filters. They posted fake news, forcing the Facebook team to come back to human editors.

    This is just one example of how security lags behind performance when humans rely on AI too much. Artificial intelligence is a great tool, but in most cases, its abilities only give reliable and the most accurate results in collaboration with eye care professionals. Although machines are designed by humans, they often can’t predict human behavior and don’t know how to cope with situations or clinical cases that go beyond the scope of the algorithm.

    Therefore AI is not dangerous for humans when ophthalmologists and optometrists periodically control the work of algorithms and review how the machine works. This is the number-two reason why artificial intelligence replace ophthalmologists and optometrists is unrealistic.

    Will AI ever be 100% objective?

    artificial intelligence replace ophthalmologists

    To honestly answer the questions of will artificial intelligence replace ophthalmologists and optometrists and whether it is 100% objective, you need to understand that an AI system will only be as good as its inputs. By loading unbiased training datasets, engineers can create an AI system that makes unbiased decisions. However, in the real world, AI is unlikely ever to be 100% objective. 

    For example, many well-known companies, such as Amazon or Facebook, still struggle with the gender gap in hiring. Some time ago, Amazon used historical data from the past ten years to train its AI recruiting model. The algorithm was supposed to process data and candidates and free recruiters from the routine viewing of hundreds of CVs. However, soon Amazon team discovered that the data was biased against women. AI algorithm was trained by outdated information when the technology industry used to be dominated by men. Thus, the new recruitment system selected only male candidates. This forced Amazon to abandon the algorithm and re-open many recruiter positions.

    In the field of ophthalmology, AI models can already accurately predict diabetes risk factors or potential vision loss from OCT images. So when will artificial intelligence replace ophthalmologists? In Altris, we are sure the algorithm will never achieve adequate objectivity, as it will always be limited by input data, whether demographics, gender, or age. 

    Now we know that AI can’t be 100% objective. Indeed, ophthalmologists and optometrists can’t match the ability of algorithms to detect pixel-level patterns among the millions of pixels in the OCT scan. However, only the cooperation of eye care specialists and a quality AI model working together will allow for more accurate detection of diseases. The combined efforts of AI management systems and eye care specialists can help achieve the desired 100%.

    Can AI make it without eye care specialists?

    artificial intelligence replace ophthalmologists

    Various articles have speculated on whether artificial intelligence replace ophthalmologists and optometrists, raising concerns about unemployment. However, this never corresponded to the actual state of affairs. Carl Benedikt Frey, an Oxford Martin Citi Fellow at Oxford University, reported that while 47% of jobs are at risk of automation, the risk for doctors is estimated at only 0.4%.

    In addition, in his book “Humans Are Underrated”, Geoff Colvin states that the most valuable skill for ophthalmologists is the ability to sense the thoughts and feelings of patients who are losing sight.

    Many patients complain about the lack of contact with the doctor. They admit that the treatment would be more comfortable if doctors devoted more time to live communication. This mainly applies to children and the elderly, who need a lot of attention from eye care specialists. Empathy and similar human qualities are not only an understanding of the patient’s feelings but also an adequate response to them. Thus, a future in which optometrists and ophthalmologists are replaced by AI seems senseless.

    Professor Tien Yin Wong, medical director of the Singapore National Eye Centre, claimed that AI holds great promise for retinal screening. And while AI for OCT interpretation will radically change clinical practice, the technology’s more significant impact will be to complement and enhance human capabilities rather than replace them. The field of ophthalmology demonstrates that the combined efforts of scientists and machines are more effective than either could achieve individually. 

    Artificial intelligence for OCT interpretation is just a recommendation system for an eye care specialist. Often one pathological sign, for example, Cystoid macular edema (CME), or Intraretinal fluid, can indicate many diseases, like Wet AMD, DR, DME, CRVO, and others. That is why AI is only an assistant to a doctor, especially when it comes to rare pathologies.

    All in all, AI for OCT interpretation is just a tiny part of clinical practice and can never work without humans. In order to detect the pathological signs and diagnose a disease correctly, an eye care specialist must perform different examination methods. Among these exams are visual acuity, intraocular pressure, ophthalmoscopy, and a basic patient examination, which includes anamnesis. Moreover, ophthalmologists and optometrists may also need to perform other visualization methods, like Fundus photography, FFA, or OCTA.

    Will artificial intelligence replace ophthalmologists and optometrists?

    Automation has led to a significant change in many industries, and ophthalmology is no exception. So when will AI replace eye care specialists? The answer is quite simple — AI will never replace them. It will eventually take over routine tasks, allowing the career of ophthalmologists and optometrists to advance in new and exciting directions.

    Automated interpretation of OCT scans will significantly increase the circulation of patients in ophthalmic clinics or optometry centers, which is commercially attractive. Moreover, with increasing life expectancy, and expanding the range and effectiveness of treatment options offered, a collaborative effort between ophthalmologists and AI will improve patient outcomes. This will make ECPs more efficient, freeing up time for human interaction between doctor and patient, which has been a cornerstone of medicine for decades.

    In his concept of the future clinic, Eric Topol describes a system that the Altris AI team is already implementing today. AI labels, annotates and segments images. While ophthalmologists receive information about the structural and functional trends of the patient’s retina to track changes and develop a treatment plan. Altris AI allows ophthalmologists to focus on providing individualized care to each patient.

  • AI for Reading Centers: How it Boosts Workflow and Efficiency

    AI medical image analysis
    Mark Braddon
    05.10.2022
    7 min read

    In recent years reading centers have become an essential resource for facilitating imaging research in many fields, including clinical trials of ophthalmology drugs. And their importance will continue to grow

    Reading centers provide crucial information by evaluating images. That is why for conducting accurate clinical trials, they must hire ophthalmologists of high qualification. Moreover, to ensure consistent analysis, the materials that graders use for the research (be it fundus photographs, fluorescein angiograms, or OCT scans) must also undergo quality control. However, even such measures can’t completely exclude errors or biases.

    Meanwhile, recent developments in the field of AI medical image analysis revolutionized the approach to clinical trials, which makes it possible to boost the workflow of reading centers. AI image analysis software works with thousands of images, efficiently providing the large amount of data needed to analyze the patient’s condition. In addition, evaluating images with AI is faster, cheaper, and more effective

    In this article, we will discuss the top 5 benefits of AI medical image analysis software for reading centers and the way AI improves the image interpretation process.

    Limitations of the manual evaluating procedure

    Although several reading centers have already implemented AI for image analysis in their workflow, most organizations are far from evaluating automation and prefer classic image interpretation methods.

    AI medical image analysis

    In most reading centers, ophthalmologists manually evaluate ocular images for drug safety studies, compile the images, and perform statistical analysis of the data. Research sizes for reading centers can range from 50 images to 3000 or more, and dozen of separate sets of images can be collected per research subject. Therefore reading centers have many obstacles to a quality evaluation process and accurate results.

    • Large amount of images is hard to proceed

    The vast number of images that need to be processed in the short term usually leads to the main problem for reading centers — most hire outsourced ophthalmologists to speed up the image grading and evaluation process. Outsourced specialists have different levels of qualification and different evaluating methods, which may lead to decreased accuracy. In addition, outsourced eye care specialists are not always interested in performing the work at the highest level. 

    • Human resources are expensive

    Another limitation of the standard evaluating procedure is the high сost spent on ophthalmologists. Human resources are usually quite expensive and associated with the risk of staff turnover. 

    • High probability of human bias

    Besides, hours spent in front of a computer screen evaluating thousands of images create a stressful environment for ophthalmologists and cause many errors, affecting the accuracy of the clinical trials. Even the FDA recognizes grader fatigue and its impact on potential errors in image interpretation. 

    • Inaccurate labeling

    In addition, administrative problems also occur quite often. This happens due to deviations from study protocols and incorrect labeling of images, which can compromise the integrity of the analyses.

    Fortunately, the pace of digitalization in reading centers is accelerating. Here is how AI medical image analysis can help reading centers cope with the growing workload. 

    The importance of implementing AI medical image analysis for reading centers

    Usually, AI image analysis is made through a pattern recognition process that involves scanning images for specific pathological signs to interpret the patient’s condition. The AI image analysis software has precise and efficient evaluation protocols that allow the analysis and interpretation of images in terms of a variety of qualitative morphological parameters. For example, when analyzing images of a patient with diabetic retinopathy, the AI models recognize microaneurysms or hemorrhages.

    AI medical image analysis

    AI algorithms allow reading centers to conduct trials of any size and duration, including various treatments for various eye diseases. Moreover, unlike the standard image interpretation process, which requires significant human resources, the introduction of AI for image analysis into the workflow of reading centers has many advantages. 

    • Quality control. Using AI algorithms ensures no errors in OCT scan analysis. AI image analysis software ensures that the desired parameters are classified based on certified imaging protocols.
    • Less money spent. Implementing AI-assisted OCT analysis is less expensive than hiring outsourced ophthalmologists. 
    • Accurate quantification. AI in medical image analysis does not depend on patient characteristics or treatment group assignment knowledge, so the machine provides the most objective and accurate assessment possible.
    • Increased efficiency. Improving the reading centers workflow with AI provides an objective and standardized classification of images. It means that any human bias is excluded, which increases the reputation of clinical research.
    • No time wasted — no more hours spent at a computer screen. Evaluating images with AI medical image analysis provides faster and more sensitive identification of the patient’s condition, which can positively impact decision-making.

    How reading centers will benefit from automated image evaluation

    In short, image evaluation with algorithms is fast, less expensive, and more reproducible. However, many companies that perform clinical trials in cooperation with reading centers are still afraid of implementing AI in medical image analysis and evaluating processes. Modern AI-based image management systems, such as Altris AI, unlike their predecessors, allow reading centers to overcome the challenges of the manual image interpretation process.

    A lot of data available to train an algorithm 

    The more images with various pathological features the algorithm has for training, the more accurately it will detect the diagnosis. Modern AI image analysis software has the ability to obtain thousands of OCT images from different models of devices for comprehensive and correct training of algorithms. Although many medical centers keep their clinical practice confidential, many ophthalmic cases and images with various pathological signs in the public domain allow the training of AI algorithms.  

    For example, the Altris AI medical image analysis software was trained on 5 million unique OCT scans obtained in 11 practicing ophthalmology clinics through the years. Our retina experts took a responsible approach to annotating and labeling images for algorithm training. A thorough error detection and correction procedure gave our algorithm 91% accuracy. 

    Constant quality control

    The responsibilities of the modern algorithms developers include not only the release of the model but also further diagnostics, which allows avoiding the problem of reproducibility. After all, constant quality control is necessary for algorithm development environments. Understanding the importance of quality control, the Altris AI team constantly tests the reproducibility of AI medical image analysis model diagnostics.

    Collection of rare diseases

    According to our research, ​25% of ophthalmologists, on average, miss rare pathologies 3 times a week.​ However, modern AI image analysis software allows overcoming this challenge. For example, Altris AI excludes missing minor, early, rare pathologies. Our team created an algorithm that automates the detection of 54 pathological signs and 49 pathologies.

    High percentage of algorithmic bias is avoided

    Algorithmic bias is one of the biggest challenges in AI. Although algorithms themselves do not have biases, they inherit them from humans. However, today, AI for image analysis has learned how to overcome the lack of interoperability between medical record systems. 

    Although it is impossible to avoid algorithmic bias completely, as it can appear at any stage of the algorithm creation process, from study design and data collection to algorithm development and model selection, modern developers take a direction to fair AI. By using a technical and regulatory framework that provides the diverse data needed to train AI algorithms, the Altris AI team makes modern technologies inclusive and ensures algorithmic bias can be excluded.

    The future of AI medical image analysis in reading centers

    The ultimate goal of the ophthalmic AI system for reading centers is to improve the grading and evaluating process and obtain more accurate research results. However, instead of fully digitalizing image assessment, the ideal approach to analysis is integration — where the benefits of AI algorithms and human skills can be combined.

    Technology will never fully replace humans, but it is already improving their work efficiency. For example, by taking over more routine and monotonous tasks, algorithms allow ophthalmologists to focus on specific eye areas and increase the evaluation speed. AI medical image analysis software can also be effective in determining compliance with the standardization of feature interpretation and determining image quality for requesting more images. 

    There are undoubtedly many challenges to integrating AI for image analysis into the workflow of reading centers. However, modern AI technologies can already overcome almost all of them. Altris AI image interpretation system is changing the future of clinical research by helping to classify images faster and increasing the efficiency, accuracy, and reproducibility of clinical trial data.

  • The Role of AI Image Interpretation for Ocular Pathologies Detection

    The use of AI for image analysis
    Maria Znamenska
    28.09.2022
    20 min read

    The burden of timely diagnostics lies on the shoulders of eye care specialists: ophthalmologists and optometrists worldwide. According to the International Agency for the Prevention of Blindness, over 1 billion people live with preventable blindness because they can’t access the proper diagnostics and treatment. Almost everyone needs access to eye care services during their lifetime. Unfortunately, there are only 331K optometrists worldwide, while 14M optometrists are required to provide effective and adequate eye care services. 

    Test Artificial Intelliegence for Automated OCT Analysis  Try for Free

     

    With the high prevalence of the population that needs eye care services and the lack of specialists, the goal of timely and accurate diagnostics and treatment seems unachievable. 

    However, the empowerment of eye care specialists with Artificial Intelligence (AI) can be a real solution to this problem. As the larger part of the work of eye care specialists relies on retina image assessment and analysis, the support of this process can unburden ophthalmologists and optometrists all over the world. Modern AI image interpretation algorithms, such as Altris AI, can discover patterns among millions of pixels with high speed, accuracy, and zero human errors because of tiredness. 

    In this article, we will talk about the capabilities of AI image interpretation for Optical Coherence Tomography (OCT) in detecting common pathologies, such as AMD or glaucoma, and less prevalent, such as Choroidal Melanoma. Despite the skepticism of the eye care community towards AI, multiple research works mentioned in this article prove the efficiency of AI. Moreover, there are market tools, capable of detecting 49 eye pathologies with 91% accumulative accuracy. Altris, the SaaS created by a team of retina experts based on 5 million OCT scans obtained in 11 clinics, is such a tool. 

     

    AI for Asteroid Hyalosis

    Asteroid hyalosis is a clinical condition in which calcium-lipid complexes are suspended throughout the vitreous collagen fibrils. Although it is a rare disease (​​1.2% prevalence according to the U.S. Beaver Dam Eye Study), it also may lead to unpleasant consequences, such as surface calcifications of intraocular lenses. Today OCT can help with the detection of this degenerative condition. For a higher confidence level, eye care specialists may use Altris AI image interpretation for OCT analysis to detect asteroid hyalosis.

    AI for Central Retinal Artery Occlusion (CRAO)

    Central Retinal Artery Occlusion (CRAO) presents as unilateral, acute, persistent, painless vision loss. It can be bilateral in 2% of the population. The vision loss is abrupt, and the treatment is only effective during the first hours. CRAO resembles a cerebral stroke. Therefore, its treatment should be similar to any acute event treatment: detecting the occlusion site and ensuring it won’t occur again. AI image interpretation models, such as Altris AI, can assist eye care specialists in detecting CRAO today. 

    AI for Central Retinal Vein Occlusion (RVO)

    AI for Central Retinal Vein Occlusion (RVO)

    CRVO is one of the most widespread vascular diseases that affect the population over 45. There are two distinct types of CRVO: perfused (nonischemic) and nonperfused (ischemic). Each of these types has its symptoms and treatment prognosis. For instance, ischemic CRVO leads to sudden visual impairment, while nonischemic CRVO development takes time to develop mildly. The detection of CRVO is now done with the help of OCT predominately, and AI image interpretation systems shows promising results in spotting its symptoms, such as nonperfusion. Altris AI system defines CRVO with 91+% accumulative accuracy in detecting pathological signs that indicate the CRVO.

    AI for Central Serous Chorioretinopathy (CSC)

    Accumulation of fluid under the central retina is called central serous chorioretinopathy. Over time, this disease can lead to the distortion of vision. Fortunately, available AI models for OCT scan analysis show high accuracy in detecting CSC. This and other AI image interpretation models effectively discriminate between acute and chronic CSC, and their performance can be comparable to the performance of ophthalmologists. Altris AI is already helping eye care specialists worldwide to diagnose CSC cases.

    AI for Chorioretinal Scar

    Chorioretinal scars are tiny scars in the back of the eye, the size of which may vary from 0,5mm to 2mm. In most cases, the chorioretinal scar appears as the result of virus infection, such as toxoplasmosis and toxocariasis, or trauma. It usually has no malignant potential. Modern AI image interpretation algorithms allow ophthalmologists and optometrists to diagnose chorioretinal scars more accurately by relying on OCT images.

    AI for Chorioretinitis

    The inflammation of the choroid is called chorioretinitis. Often, the inflammatory process can be caused by congenital viral, bacterial, or protozoan infections. Chorioretinitis is characterized by vitreous haze, fine punctate gray to yellow exudation areas, pigment accumulation along the optic nerve and blood vessels, and flame-shaped hemorrhages with chorioretinal edema. The goal of the eye care specialist is to detect chorioretinitis which can potentially lead to blindness, and to eliminate inflammation. Altris AI image interpretation system can be an excellent decision-making support tool in detecting chorioretinitis.

    AI for Choroidal Melanoma

    Today, choroidal melanoma is the second most common intraocular tumor in the adult population. Patients with choroidal melanoma don’t have distinct symptoms but can have impaired visual acuity, visual field defects (scotomas), metamorphopsia, photopsia, and floaters.

    OCT is a relatively new method for the detection of choroidal melanoma, which is nevertheless gaining popularity. OCT cannot be the only diagnostic method for melanoma detection – FA is also needed for final diagnosis. However, optical shadowing, thinning of overlying choriocapillaris, subretinal fluid, retina local elevation, subretinal lipofuscin deposits, and disrupted photoreceptors can be detected with the help of OCT.

    Such pathological signs will indicate possible choroidal melanoma. Altris AI image interpretation system can assist eye care specialists with detecting pathological b-scans and locating this disease.

    AI for Choroidal Neovascularization (CNV)

    AI for Choroidal Neovascularization (CNV)

    Choroidal neovascularization (CNV) is part of the spectrum of exudative age-related macular degeneration (AMD) and some other conditions. CNV is an abnormal growth of vessels from the choroidal vasculature to the neurosensory retina through Bruch’s membrane.

    Modern OCT systems can detect even a tiny amount of fluid leaking into the retina. Empowered by Altris AI image interpretation algorithm, eye care specialists can spot pathological signs of choroidal neovascularization much faster or detect the pathologies that accompany CNV, resulting in better patient outcomes.

    AI for Choroidal Rupture

    Traumatic choroidal rupture is common after blunt ocular trauma (5 to 10%). It is a defect in the Bruch membrane, the choroid, and the retinal pigment epithelium. The location of the choroidal rupture will define the symptoms: if the fovea and parafoveal retina are included in the rupture area, patients experience impaired vision. In other cases, the rupture can be asymptomatic. OCT is used to diagnose choroidal rupture as it can show the loss of continuity of the RPE layer and the thinning of the choroid. AI image interpretation is exceptionally accurate in layers segmentation and volume/area calculation, so missing the symptom of choroidal rupture with AI is almost impossible.

    AI for Choroidal Nevus

    AI for Choroidal Nevus

    Choroidal nevus is a benign melanocytic tumor of the choroid and is found in 5 to 30% of white people. It can be found accidentally because it is asymptomatic. Artificial intelligence methods are used not only for identifying choroidal nevus but also for early signs of its transformation into malignant melanoma. The earlier the small melanoma is detected, the better the treatment prognosis is for the patient. Altris AI image interpretation system is one of the systems capable of detecting choroidal nevus before its transformation into melanoma. 

    AI for Cone-Rod Dystrophy (CORD)

    CORD is an inherited retinal disease caused by a genetic mutation characterized by cone photoreceptor degeneration. It may be followed by subsequent rod photoreceptor loss. CORD symptoms include loss of central vision, photophobia, and progressive loss of colored vision. OCT diagnostics help to diagnose CORD by pointing at the absent interdigitation zone and progressive disruption and loss of the ellipsoid zone (EZ). Today AI image interpretation is helping to detect Cone-Rod Dystropthy to eye care specialists more confidently, even in controversial cases.

    AI for Cystoid Macular Edema (СME)

    AI for Cystoid Macular Edema (СME)

    Cystoid macular edema (CME) is a painless condition in which cystic swelling or thickening occurs of the central retina (macula) and is usually associated with blurred or distorted vision. CME can be caused by many factors, including diabetic retinopathy and age-related macular degeneration (AMD). OCT diagnostics help to spot СME by detecting retinal thickening with the depiction of the intraretinal cystic areas. CME is not irreversible. Vision loss caused by macular edema can be reversed if detected early. Combining OCT diagnostics with AI image interpretation, eye care specialists can detect CME with higher accuracy at an earlier stage.

    AI for Degenerative Myopia

    AI for Degenerative Myopia

    Degenerative or pathological myopia is the condition during which axial lengthening occurs, especially in the posterior pole. It leads to retina stretching, the sclera’s thinning, choroidal degeneration, and potential loss of vision. AI image interpretation systems have demonstrated excellent results in detecting pathologic myopia and identifying myopia-associated complications on OCT. AI helps ophthalmologists improve the monitoring of pathology treatment and classify different cases of myopia.

    AI for Diabetic Macular Edema

    ​​Diabetic macular edema (DME) is the presence of excess fluid in the extracellular space within the retina in the macular area, typically in the inner nuclear, outer plexiform, Henle’s fiber layer, and subretinal space. DME can develop during any stage of diabetic retinopathy in patients with diabetes.

    Unfortunately, the early symptoms of DME can be unnoticeable or include impaired vision and reading and color perception problems, which some people may ignore. Taking into account its asymptomatic nature, patients with diabetes need regular OCT examinations to determine the presence of DME. OCT has become a golden standard in DME detection within the last few years, and AI can be an excellent decision-making support tool in OCT scans interpretation. According to recent research, AI-powered OCT analysis provides an accurate diagnosis of DME with a cumulative accuracy of over 92%. 

    AI image interpretation systems can unburden ophthalmologists and optometrists who have a lot of patients due to their convenience and can be used in remote regions of the world in the future.

    AI for Diabetic Retinopathy

    AI for Diabetic Retinopathy

    Diabetes can affect the eyes in various ways, most commonly corneal abnormalities, glaucoma, iris neovascularization, cataracts, and neuropathies. However, diabetic retinopathy (DR) is the most common and potentially the most blinding of these complications. Early treatment of both proliferative and non-proliferative DR can improve patient outcomes significantly. OCT is a common diagnostic method for diabetic retinopathy. It relies on the localization of intraretinal and/or subretinal fluid and can help to diagnose diabetic retinopathy through pathological signs detection and layer thickness measurement. 

    AI image interpretation is a step in the future of detection that shows high sensitivity in identifying DR, and studies prove its effectiveness. AI-assisted analysis of OCT scans helps eye care specialists today and will definitely be more widespread tomorrow.

    AI for Dry AMD

    Dry AMD is a more common type of AMD (80% of people have this type), during which patients slowly lose their central vision. It is the aging of the macula and the appearance of deposits called drusen. There is no treatment for Dry AMD yet. However, early detection can help patients to change their lifestyles and slow down the development of this disease. Modern AI solutions make it possible to diagnose Dry AMD faster and develop successful methods of treating the disease. AI image interpretation systems also exclude the possibility of human error.

    AI for Dry AMD – Geographic Atrophy

    Geographic atrophy is an advanced form of the late stage of Dry AMD development. In this condition, retina cells will degenerate and finally die, leading to the patient’s central vision loss.

    AI image interpretation is being widely used to detect Geographic Atrophy with the help of OCT. In this meta research, there are numerous studies that focus on lesion segmentation, detection, and classification of geographic atrophy and even its prediction. They vary in accuracy, but the overall trend of AI for geographic atrophy detection is very positive. The use of artificial intelligence has several advantages, including improved diagnostic accuracy and higher processing speed. 

    AI for ERM or Epiretinal Fibrosis

    AI for ERM or Epiretinal Fibrosis

    Epiretinal fibrosis (epiretinal membrane or macular puckering) is a treatable cause of visual impairment. It is a macula disease caused by fibrous tissue growth on the retina surface. AI image interpretation model for detecting ERM on OCT can outperform non-retinal eye care specialists with a cumulative accuracy of 98+%. For more professional retina experts, AI can be a decision-support tool. Early detection and treatment of this disease are crucial to prevent the ​​growth of fibrous tissue and the worsening of the patient’s condition.

    AI for Epiretinal Hemorrhage

    Epiretinal hemorrhages result from a serious trauma: car or sports accidents, falls, and direct physical impact. Mild hemorrhages unrelated to a serious traumatic event can disappear on their own, but they can be a symptom of a more complex pathology. Epiretinal hemorrhages can be detected with the help of OCT, and AI image interpretation systems can make this process more accurate.

    AI for MTM ( Foveoschisis)

    Myopic foveoschisis or myopic traction maculopathy is the thickening of the retina that reminds schisis in patients with high myopia with posterior staphyloma.

    Untreated foveoschisis often leads to vision loss due to secondary complications, which is why this disease should be detected in time. Today when OCT is becoming more widespread, detection of foveoschisis is more common and accurate. More than that, the studies show that combining the power of AI image interpretation and OCT diagnostics for MTM detection is equal to the junior ophthalmologist’s knowledge. Using AI-powered OCT, it is possible to deal with the shortage of specialists that can guarantee timely diagnostics.

    AI for Full-thickness Macular Hole

    AI for Full-thickness Macular Hole

    A macular hole is a full-thickness defect of the retina involving the foveal region. Patients usually present a reduction of central visual acuity. A complete ophthalmic examination, including OCT, should be performed to diagnose a full-thickness macular hole. So far, the research of AI image interpretation algorithms for a full-thickness macular hole is dedicated to OCT(A), but there are available tools on the market that can help define full-thickness macular hole on OCT scans as well. Altris AI is one of them.

    AI for Hypertensive Retinopathy

    People with high blood pressure, older people, and patients with diabetes often develop hypertensive retinopathy. OCT examination can be used for the detection of hypertensive retinopathy.

    AI-based OCT analysis shows promising results in detecting hypertensive retinopathy by defining retinal vessels and other pathological signs in the retina.

    AI for Intraretinal Hemorrhage

    AI for Intraretinal Hemorrhage

    Among patients with DR, RVO, or ocular ischemic syndrome, there are often those who develop side pathologies. One of these pathologies is intraretinal hemorrhage. AI image interpretation systems help ophthalmologists and optometrists identify intraretinal hemorrhages in the retina.

    AI for Vitreous Hemorrhage

    Vitreous hemorrhage results from bleeding into one of the several potential spaces formed around and within the vitreous body. This condition can follow injuries to the retina and uveal tract and their associated vascular structures. Eye care specialists should perform a complete eye examination, including OCT, slit lamp examination, intraocular pressure measurement, and dilated fundus evaluation. Timely diagnosis and treatment are essential: it can significantly reduce concomitant diseases of intravitreal hemorrhage. AI image interpretation systems can help eye care specialists detect vitreous hemorrhage supporting them in case of controversial OCT scans.

    AI for Lamellar Macular Hole (LMH)

    AI for Lamellar Macular Hole (LMH)

    Lamellar macular hole is one of the types of macular holes known in eye care practice. The problem is that the stage 0 macular hole is a clinically silent finding detected on OCT where a parafoveal posterior hyaloid separation is present and a minimally reflective preretinal band is obliquely inserted at one end of the fovea. Eye care specialists may have problems identifying lamellar macular hole on OCT. That is where AI image interpretation models can come into play.

    AI for Laser-induced Maculopathy

    Since 2014, the number of laser injuries reported worldwide has more than doubled because of the widespread use of laser technologies. Depending on the damage, the patient may have a quick recovery or long-term vision loss with the development of diseases such as photoreceptor’s damage, macular hole, ERM, or others. OCT is one of the methods that help to detect laser-induced maculopathy without human errors and doubts. AI image interpretation models have a reasonable prospect of helping eye care specialists define laser-induced maculopathy based on OCT scans.

    AI for Age-related Macular Degeneration (ARMD)

    Age-related macular degeneration is one of the leading reasons for blindness in people of older age, especially among women and people with obesity. Patients usually present with a gradual, painless vision loss associated with delayed dark adaptation, severe metamorphopsia, and field loss. In other words, in the early stages of AMD, patients may not have any signs or symptoms, so they may not even know they have the disease. Regular OCT screening (among other diagnostic methods) can be a life-saving vest for older people.

    AI image interpretation has shown great promise in detecting AMD, and research papers show that its capabilities are similar to those of ophthalmologists. AI-powered automated tools provide significant benefits for AMD screening and diagnosis.

    AI for Macular Telangiectasia Type 2

    Macular telangiectasia (Mac Tel) results from the capillaries abnormalities of the fovea or perifoveal region related to the retina nuclear layers and ellipsoid zone.

    Macular Telangiectasia Type 2 can have negative consequences and develop into cystic cavitation-like changes in all the layers of the retina or even transform into a full-thickness macular hole. OCT is an effective diagnostic method of macular telangiectasia type 2 as the tomograph can localize foveal pit enlargement. Which is a result of secondary loss of the outer nuclear layer and ellipsoid zone that can progress into large cysts (often called ‘cavitation’) that can encompass all retinal layers.

    Automating the detection of macular telangiectasia type 2 with the help of AI image interpretation systems for OCT scan analysis is already possible thanks to Altris AI.

    AI for Myelinated Retinal Nerve Fiber Layer

    Myelinated nerve fiber layer (MRNF) is a disease that occurs in 1%. It is a benign clinical condition that results from an embryologic developmental anomaly whereby focal areas of the retinal nerve fiber layer fail to lose their myelin sheath.

    OCT is an effective method of MRNF detection with the help of the detection of the RNFL layer. Such tools as Altris AI image interpretation models are even more accurate in retina layers detection and volume measurement thanks to their growing level of accuracy.

    AI for Myopia

    AI for Myopia

    Myopia is not an eye disease. It is an eye-focusing disorder that affects 25% of the world population at a younger age. There are 2 distinct types of myopia: pathological and non-pathological — each of the types has its symptoms and treatment prognosis. The visual function of the patients, as well as the high quality of life, can be preserved if myopia is detected early enough and treated appropriately. Myopia is often diagnosed by ophthalmologists and optometrists with the help of OCT, thanks to its fine cross-sectional imagery of retinal structures. Unlike biomicroscopy, angiography, or ultrasonography, OCT can reveal undetectable retinal changes in asymptomatic patients with myopia.

    Current AI image interpretation models show great promise in detecting myopia on OCT scans, and their results can be compared to the results of junior retina specialists. Altris AI is an accurate AI tool for myopia.

    AI for Pigment Epithelium Detachment

    Retinal pigment epithelial detachment (PED) is often observed in Wet AMD and other conditions. It is determined as a separation of the RPE layer from the inner collagenous layer of Bruch’s membrane. With its capability to visualize retinal layers, OCT helps eye care specialists with timely PED diagnostics. Powered with AI image interpretation systems, OCT diagnostics can promise zero human errors and exceptional accuracy.

    AI for Polypoidal Choroidal Vasculopathy (PCV)

    Polypoid Choroidal Vasculopathy is a disease of the choroidal vasculature. Serosanguineous detachments of the pigmented epithelium and exudative changes that can commonly lead to subretinal fibrosis are the main OCT signs of PCV. AI image interpretation systems show great potential in establishing a difference in diagnostics between PCV and AMD.

    AI for Preretinal Hemorrhage

    AI for Preretinal Hemorrhage

    Preretinal hemorrhage is a complication of many pathologies, such as leukemia or ocular/head trauma. Missing preretinal hemorrhage means putting a patient at risk. Preretinal hemorrhage can be a presenting sign of some systemic diseases. In any case, OCT diagnostics are performed to determine preretinal hemorrhage and its real reason.

    AI for Pseudohole

    Sometimes the pulling or wrinkling of the epiretinal membrane (ERM) can result in a gap called a pseudohole. A pseudohole can look like a macular hole; sometimes, it can turn into one, so it is essential to distinguish between these two phenomena. Optical coherence tomography can accurately determine a pseudohole revealing an epiretinal membrane with contraction of the retina or suppression of retinal layers. Combined with AI image interpretation, OCT diagnosis can guarantee higher accuracy in pseudohole detection.

    AI for Retinal Angiomatous Proliferation (RAP)

    RAP is a subtype of AMD, which is neovascularization that starts at the retina and progresses posteriorly into subretinal space. There are 3 stages of RAP: intraretinal neovascularization (IRN), subretinal neovascularization (SRN), and choroidal neovascularization (CNV). OCT is effective for detecting IRN only since changes beneath the pigment epithelium are challenging to assess. AI image interpretation models effectively differentiate between RAP and polypoidal choroidal vasculopathy (PCV), comparable to the performance of eight ophthalmologists. AI cannot substitute eye care specialists but can be an excellent decision-making support tool.

    AI for Retinal Detachment

    Retinal detachment is a serious eye condition that happens when the retina pulls away from the tissue around it. It can be a result of trauma or another disease. OCT has become a new standard for detecting early retinal detachment and defining the best time for surgical operation, for example. OCT powered with AI image interpretation systems can give eye care specialists the confidence they need to determine the degree of detachment and make the correct prognosis.

    AI for Retinitis Pigmentosa

    RP is a hereditary diffuse pigment retinal dystrophy characterized by the absence of inflammation, progressive field loss, and abnormal ERG. OCT diagnostics allows assessing morphological abnormalities in RP, providing insights into the pathology of RP and helping to make a good prognosis. AI image interpretation applied for the OCT analysis shows promising results in Inherited Retinal Diseases detection and future management.

    AI for Retinoschisis

    AI for Retinoschisis

    Retinoschisis is an eye condition characterized by a peripheral splitting of retinal layers. OCT is an effective method of retinoschisis diagnostic. The application of AI image interpretation tools for OCT analysis for identifying retinoschisis (among other myopia conditions) is comparable to the performance of experienced ophthalmologists.

    AI for Retinal Pigment Epithelial (RPE) Tears (Rupture)

    RPE rupture or RPE tears is the condition when this retinal layer acutely tears from itself and retracts in an area of a retina, usually overlying a pigment epithelial detachment (PED). OCT is an effective method of diagnostics of RPE tears. OCT scans will show a discontinuity of the hyperreflective RPE band, with a free edge of RPE usually wavy and scrolled up overlying the PED, contracted back towards the CNVM.

    AI image interpretation systems can provide eye care specialists with confidence when detecting RPE tears. Systems such as Altris AI can distinguish between retinal layers with exceptional accuracy, exceeding the accuracy of eye care specialists.

    AI for Solar Retinopathy (Maculopathy)

    Solar retinopathy is photochemical toxicity and the consequent injury to retinal tissues located in the fovea in most cases. OCT helps to diagnose solar retinopathy by indicating changes and focal disruption at the level of the subfoveal RPE and outer retinal bands. The overall retinal architecture remains intact. AI image interpretation models can confidently assist eye care specialists in detecting solar retinopathy, even when they are in doubt.

    AI for Subhyaloid Hemorrhage

    Subhyaloid hemorrhage is diagnosed when the vitreous is detached from the retina because of blood accumulation. This type of hemorrhage is rare and is different from intraretinal hemorrhage caused by trauma or diabetes. OCT helps to detect subhyaloid hemorrhage. For eye care specialists who don’t have experience in detecting subhyaloid hemorrhage, the AI image interpretation model can become a great support tool.

    AI for Subretinal Fibrosis

    Subretinal fibrosis appears due to wound healing reaction to the choroidal neovascularization in nAMD or other conditions. Early diagnostics of subretinal fibrosis are critical because a neovascular lesion’s transformation into a fibrotic lesion can be very rapid. OCT is regarded as the most accurate method of diagnostics today.

    AI image interpretation systems can help eye care specialists who use OCT with early diagnostics of subretinal fibrosis and improve patient outcomes.

    AI for Subretinal Hemorrhage

    AI for Subretinal Hemorrhage

    Subretinal hemorrhages are a complication of various diseases which arise from the choroidal or retinal circulation. It is most often caused by AMD, trauma, and retinal arterial macroaneurysm. OCT will be an effective tool for determining the level at which subretinal hemorrhage occurred. Powered with AI image interpretation models, OCT can become the decision-making support tool eye care specialists need for subretinal hemorrhage identification.

    AI for Sub-RPE (Retinal Pigment Epithelial) Hemorrhage

    Sub-RPE (retinal pigment epithelium) hemorrhage is located between the RPE and Bruch’s membrane. OCT is an essential tool for validating the hemorrhage’s diagnosis and localization. AI image interpretation tools, such as Altris AI, will ensure that Sub-RPE hemorrhage is not missed.

    AI for Tapetoretinal degeneration or dystrophy

    Tapetoretinal dystrophy or tapetoretinal degeneration (TD) is exogenous destruction of the retina caused by a genetic mutation. Eye care specialists might easily miss such rare conditions as tapetoretinal degeneration. It is often advisable to have AI image interpretation systems as a decision-making support tool not to miss TD or other uncommon diseases.

    AI for Vitelliform Dystrophy

    It is autosomal dominant degenerative maculopathy wherein a mutation in the bestrophin gene leads to lipofuscin accumulation in RPE cells manifested in a yellow spot. Detecting vitelliform dystrophy is critical at the early stages as it can lead to vision loss. OCT provides essential information on the lesion’s morphology, location, and dynamics. Empowered with AI image interpretation tools, such as Altris AI, eye care specialists won’t miss such a rare disease as vitelliform dystrophy at the early stage.

    AI for Vitreomacular Traction Syndrome

    Vitreomacular traction syndrome is a pathological condition characterized by a posterior vitreous detachment that leads to blurred vision or serious vision impairment. OCT is an essential method of diagnostics of vitreomacular traction syndrome as it can show the amount of involvement and tension on the macula caused by VMT. Combined with AI image interpretation tools, OCT analysis can give incredible results.

    AI for Wet AMD

    AI for Wet AMD

    Wet AMD is the most widespread disease among the elderly population in developing countries. It is a disease characterized by abnormal blood vessel growth under the retina. Understanding that this disease can lead to rapid and severe vision loss, its early detection and treatment are very important. OCT is a golden standard for the diagnostics of wet AMD as it shows fluid or blood underneath the retina without dye, among other pathological signs.

    Today AI shows promising results in predicting the development of wet AMD based on OCT images. For instance, the DARC algorithm designed for detecting apoptosing retinal cells could predict new wet-AMD activity. Another effective AI image interpretation algorithm determines the location and volumetric information of macular fluid within different tissue compartments in wet AMD, providing eye care specialists with the ability to predict visual acuity changes.

    AI for X-linked Juvenile Retinoschisis (XLRS)

    XLRS is a rare congenital retina disease caused by mutations in the RS1 gene, which encodes retinoschisin, a protein involved in intercellular adhesion and likely retinal cellular organization. The disease usually affects younger males in their teenage years who complain about blurred vision. OCT is used to detect schisis in the superficial neural retina and thinning of the retina. Despite the lack of research articles on AI in OCT diagnostics of XLRS, there are AI image interpretation tools that already cope with this task effectively.

    Final Words

    Artificial intelligence can identify, localize, and quantify pathological signs in almost every disease of the macula and retina. That is how AI image interpretation systems can provide decision-making support with the pathologies at their early stages or rare pathologies. AI can help to detect many pathologies that are invisible to the human eye because of their size or that are at their early stage. 

    The overall potential of artificial intelligence for ophthalmologists and optometrists is enormous and includes pathological scan selection and scan analysis with the probability of existing pathologies and pathological signs. One trial is worth a thousand words in the case of AI tools for ophthalmologists and optometrists.

  • Types of Optometry Eye Examination Techniques & the Role of OCT

    oct scan interpretation
    Mark Braddon
    14.09.2022
    7 min. read

    Various optometry eye examination techniques have always played a crucial role in diagnosing many eye diseases and promptly referring to a retinal expert. According to Essilor International research, poor vision is the most common disability in the world today. The good news is that 90% of vision loss cases are treatable or preventable if discovered in their early stages.

    However, by performing only traditional optometric eye examination techniques, such as anterior and posterior segment examinations, optometrists may miss the complete picture of a patient’s eyes. That is why optometry specialists are embracing a new technique: optical coherence tomography (OCT) examination. 

    Test Artificial Intelliegence for Automated OCT Analysis  Try for Free

    OCT imaging helps go beyond standard eye examination procedure by better visualizing the eye’s structures and providing an additional quantitative assessment.

    In this article, I will discuss the most important optometric examination and emphasize the role of OCT scan interpretation in optometry.

    Types of optometry eye examination techniques

    When performing a full optometric examination, the optometrist should not only assess the visual acuity with an eye chart but also check their eye health. The types of eye examination tools are now very diverse and depend on the application field and the qualification level. Nowadays, there are a few eye examination techniques, although they may vary from country to country, that help diagnose a patient more accurately and improve follow-up care.

    Ophthalmoscope examination

    eye examination techniques

    Ophthalmoscopy plays a crucial role in detecting the conditions of the retina, blood vessels, and optic disc. This is a basic eye examination procedure that optometrists usually perform to evaluate many diseases, such as diabetic retinopathy or retinal vein occlusion. 

    During the direct ophthalmoscopy, the optometrist shines a light into the patient’s eyes to see the inside. Binocular indirect ophthalmoscopy also involves shining a light into the patient’s eyes, however, it allows eye care specialists to take a better look at the retina and its parts that are difficult to see with other eye examination techniques. The indirect ophthalmoscopy is usually combined with pupil dilation and another optometry practice called scleral depression.

    Slit lamp examination

    eye examination techniques

    A slit lamp consists of a microscope, light source, and frame on which a patient lies their head. This regular eye examination procedure lets an optometrist focus on the eye by working with the light: expand or narrow it, increase brightness, and filter with colors. Sometimes the procedure also includes putting a few dye drops in a patient’s eye to examine some of its parts.

    Slit lamp examination is pain-free and allows an optometrist to view the sclera, iris, or cornea to detect diseases related to allergies, autoimmune disorders, gout, or even melanoma. Such eye examination procedure also allows to view the retina of the eye to detect the pathological signs of diabetes. Optometrists usually use a slit lamp along with an ophthalmoscope examination.

    Refraction test

    eye examination techniques

    One more type of eye examination techniques is a refraction test, usually performed to detect if a patient needs glasses or contact lenses. This test made with a phoropter is quick and painless. During the optometric examination, the optometrist adjusts the power of the lenses by moving or turning them back and forth until a patient can clearly see the letters on the chart.

    An optimal value of 20/20 is considered ideal vision, while a deviation means a refractive error. This may indicate that when light passes through the lens of the patient’s eye, it is not refracted properly. An optometrist can detect astigmatism, myopia, presbyopia, and a refractive eye problem during a refraction test. This, in turn, helps detect macular degeneration, retinal vein occlusion, retinitis pigmentosa, and retinal detachment.

    • Cycloplegic refraction

    Sometimes the optometrist may decide that the normal refraction is insufficient or inaccurate due to error. During refraction, the patient may unconsciously focus, affecting the test result and showing nearsightedness or farsightedness.

    Then the optometrist performs cycloplegic refraction using cycloplegic eye drops. This eye examination procedure paralyzes the muscles that focus the eye to determine the refractive error. Сycloplegic refraction exam is especially useful for children, patients with pre-presbyopia, and LASIK patients.

    • Autorefraction

    Autorefraction is an eye examination procedure performed using a special autorefractor device, also called an optometer. This exam automates the estimation of refraction and determines its error. Usually, the indications for the procedure are myopia, farsightedness, astigmatism, presbyopia, and prescription of glasses and contact lenses.

    Retinoscopy

    eye examination techniques

    One more optometric examination usually performed to detect farsighted, nearsighted, or astigmatism, and the need for glasses is retinoscopy. This procedure is pain-free and quick. Using a retinoscope, the optometrist projects a beam of light into the patient’s eye. This light moves along a horizontal and vertical trajectory, reflecting off the back of the eye. The eye care practitioner observes the movement of light with the help of lenses they place in front of the eye. Then the optometrist changes the lens’s power and tracks the reflection’s direction and pattern. This test is performed to find a possible anomaly.

    Role of OCT scan interpretation in optometric examination

    The types of optometry eye examination techniques described above are fundamental for any diagnosis. However, adopting modern OCT scan interpretation systems already complements clinical practice perfectly and has the prospect of widespread distribution among optometrists worldwide. 

    Knowing that the prevalence of some eye conditions, such as Myopia or Dry AMD, has increased with the pandemic, specialists need to implement modern methods and eye examination techniques in their clinical practice. Current optical coherence tomography devices allow optometrists to perform consistent analysis and furthermore have special software and a database for storing patient information. Compared to other retinal examination methods, such as fundus photography, OCT scan interpretation enhances patient care by improving the quality of diagnosis.

    High-quality information provided

    Modern OCT diagnostics allow optometrists to quickly obtain a huge amount of information about the patient’s eye. Built-in software collects images and compares results to normative databases. This allows optometrists to easily track patient progress or regression and generate reports that ophthalmologists or surgeons may need for follow-up treatment.

    For example, suppose a patient has a disorder with the optic nerve, macula, or vascular system. In that case, the optometrist can send data to the ophthalmologist promptly, highlight important aspects of the patient’s condition, and provide abnormal OCT scan results for additional clarity. 

    No missed pathologies

    OCT provides higher diagnostic standards, ensuring fewer pathologies or pathological signs are missed. OCT scan interpretation helps detect early vision-threatening eye conditions. For example, the system can detect AMD in the early stages, which is crucial for preventing vision loss due to subretinal fibrosis. With OCT, the thickness of the retina over the macula and posterior pole can be analyzed to detect retinal edema or atrophy. Optometrists can also confirm diabetic macular edema (DME) and decide on further treatment based on the results of its examination. In addition, OCT perfectly visualizes the retinal pigment epithelium (RPE) and choroid.

    More patients served with comfort

    By better visualization of the eye structures, optometrists provide each patient with an individual approach. This level of service ensures comfort for patients and trust for a specialist. OCT allows optometrists to avoid routine work and devote more time and energy to patients. More importantly, the OCT scan interpretation helps establish contact, allowing patients to understand the examination and treatment plan.

    Impact of AI on OCT eye examination techniques

    OCT scanning allows optometrists to accumulate large amounts of patient data. However, a large amount of information can be difficult and time-consuming to process, even for experienced specialists. The collaboration of OCT and artificial intelligence (AI) gives optometrists a unique opportunity to analyze a large amount of data and make better clinical decisions. Here are 4 key benefits of AI which completely transform the OCT scan interpretation process for optometrists:

    • Gaining confidence. 16.3% of interviewed eye care practitioners still avoid using OCT in their daily practice because of the lack of confidence in their interpretation skills. However, with AI, this problem will be solved.
    • Fast examination. Implementing AI-powered management systems in daily clinical practice reduces the time optometrists have to spend on non-pathological scans.
    • Clear diagnosis. 59% of specialists acknowledge that they have to interpret controversial scans around 1-3 times a week. AI helps optometrists with controversial and abnormal OCT scans, so they don’t need to guess the diagnosis.
    • High diagnostic standards. 30,5% of interviewed ECPs admit they are unsure how often they miss pathologies. When working with OCT, AI systems ensure no minor, early, rare pathologies are missed.

    OCT scanning allows specialists to easily, quickly, and safely obtain many images, producing a lot of data. As AI aims to work with large volumes of data, more and more AI models are being created to help optometrists.

    eye examination techniques

    Altris AI has developed an artificial intelligence platform to assist ECPs during their optometric examination and already plays a significant role in diagnosing and treating eye diseases using OCT. We have trained an AI algorithm on 5 million OCT scans collected in 11 ophthalmic clinics with a 91% accuracy.

    Future of OCT scan interpretation in optometric examination

    The integration of OCT into the clinical practice of optometrists is beneficial and shows great promise. However, to gain the most accurate diagnosis, the interpretation of scans should be carried out in cooperation with other optometry eye examination tools. Optical coherence tomography implemented with other eye examination techniques, including gonioscopy or slit lamp biomicroscopy, boosts diagnostic performance and provides valuable data.

    OCT scan interpretation in optometry practice is becoming routine for providing improved examination and patient care. This technology can also improve the confidence of eye care specialists. Detecting many pathologies using optical coherence tomography has an immediate practical benefit. Due to its high resolution, it defines and identifies early pathological signs before patients even notice any symptoms. 

  • Top 11 Optometry & Ophthalmology Mobile Apps for Eye Care Specialists

    Maria Znamenska
    15.08.2022
    10 min. read

    Today, there are hundreds of ophthalmology mobile apps available to both experienced eye care specialists and beginners. Some of them assist in learning and practice as clinical tools, and some of them are educational applications. Some mobile applications are basically a database of useful materials, ophthalmic atlases so to say. In this post, I will focus on educational optometry and ophthalmology apps and highlight their main features and functions.

    Test Artificial Intelliegence for Automated OCT Analysis  Try for Free

     

    Altris Education OCT

    mobile ophthalmology app

    Altris Education OCT is a unique free ophthalmology mobile app that contains millions of OCT scans labeled by a team of retina experts. More than 9000 eye care specialists have already joined the application.

    The app is interactive, which means that eye care specialists can highlight pathological signs on the scan 1 by 1 to learn about their location.

    The database of OCT scans is updated every day with a new labeled OCT scan, so users can gather their library right within the app. 

    Interactive eye atlas 

    The home page of the Altris Education OCT ophthalmology mobile app consists of 4 sections: 

    • In the Feed section, users will find millions of OCT scans of the retina to practice and improve their skills. 
    • In the Folders sections, there are 41 folders with various hereditary diseases, pathologies, and pathological signs. If an eye care specialist uploads the app for a specific reason, for example, to learn how to detect Epiretinal Fibrosis, he/she can easily find a folder with needed scans and work on them.
    • In the News section, users can find recent news from the OCT world and current researches.  
    • In the Community section, a user can create a post and discuss curious cases with their colleagues. 

    Community interaction

    A team of Altirs Education OCT has the aim to build a real community of ophthalmologists and optometrists worldwide who share their passion for learning. Most eye care specialists often face difficulty while interpreting OCT scans in their everyday clinical practice. We created a community where each app user can discuss problematic scans or ask OCT-related questions ( what OCT equipment to choose?). 

    Moreover, the Altris team will engage experienced OCT experts in the forums to give a professional assessment of the scans. 

    In addition, the Altris ophthalmology mobile app allows its users to like, comment and share OCT scans, as well as save them in a personal library. 

    Special features

    In Altis ophthalmology mobile app, each pathological sign is highlighted with a different color so eye care specialists can easily learn how to interpret OCT scans. Each scan contains two tabs: pathologies and diagnosis, so users are able to highlight the pathologies in the first place and then guess the diagnosis. To check himself/herself, a user switches to the diagnosis tab and finds out the name of the disease. What is more, he/she can zoom in on OCT scans to view pathological signs in detail. 

    Membership options/perks

    Altris ophthalmology mobile app not only provides its users with a huge database of educational materials. It also engages eye care specialists to invite friends, gain budges and upgrade their level. To reach the next level, there are tasks like “Search your first scan” or “Learn 5 scans in detail”. When a user level up, he/she gets access to new folders with pathological scans. 

    Another great feature of the app is that it constantly sends its users an unfamiliar OCT scan, so they can explore something new on a daily basis.

    The basic functionality of the app is completely free. However, ophthalmologists and optometrists can also become Pro users of Altris Education OCT and unlock more scans and app features for  $4 monthly or $25 annually.

    Please upload this FREE app if you are interested:

    👉 Android link: https://lnkd.in/dqy-kvVg
    👉iOS link: https://lnkd.in/dUemq-pK

     

    Eye Handbook

    mobile ophthalmology app

    Being on the market since 2010, Eye Handbook is well known and loved by many ophthalmologists and optometrists. Eye Handbook is used worldwide for both diagnosis and treatment, as the app provides eye care professionals with tools for acuity testing, children’s target fixation, or color vision testing. Now let’s take a closer look at the app’s functionality.

    Eye atlas 

    The overview of diseases in the mobile ophthalmology app begins with the Eye Atlas tab, which is a database of various pathologies arranged in alphabetical order. The description of each disease is accompanied by fundus photos, OCT images, or fluorescein angiography. Users can sort pathologies by category choosing, for example, retinal diseases, glaucoma, or oculoplastics. 

    Moreover, with the Eye Handbook ophthalmology mobile app, users can view videos of ophthalmic surgeries, such as posterior polar cataract surgery, and many more. Users are also able to sort videos by most relevant or ranked. In addition to videos, the application provides ophthalmologists and optometrists with access to audio materials, flash cards, and slides.

    Community interaction

    The Eye Handbook mobile ophthalmology app has a forum with topics open for discussion. Users can become a part of the community, add their posts, choose the appropriate category and invite like-minded eye care specialists to discuss the latest news in the field of ophthalmology. 

    Educational materials

    The Eye Handbook is a very useful application not only for ophthalmologists but also for optometrists. Not to mention a bunch of study materials, the application has collected a large number of vision tests such as Amsler grids, duo-chrome test, OKN drum, and a lot more.

    The ophthalmology mobile app contains a variety of calculators, like the Glaucoma risk calculator, which eye care specialists can use in their clinical practice right from their smartphones. Eye Handbook gathered even coding, like ICD-10 or CPT. In the app, they are also able to find detailed information about ophthalmic meds, check the EHB manual, and get access to a constantly updating news feed.

    Eye Emergency Manual

    mobile ophthalmology app

    Eye Emergency Manual mobile ophthalmology app is a great emergency aid because it quickly provides basic information about eye diseases. The application has several features, which I will explain in more detail below.

    Eye atlas

    This mobile ophthalmology app provides structured and detailed information about many eye traumas and treatments. Users can find fundus photos, photographs of real people’s eyes, or scans of each trauma and read about their initial treatment. In some cases, the developers even created Eye Trauma Communication Checklists to help eye care specialists come to a medical conclusion many times faster. 

    The Eye Emergency Manual app also contains a database of acute red eye or eyelid cases. All the information is presented clearly and plainly.

    Special features

    Each pathology overview can be saved so the app users can later explore their favorite pages or favorite glossary terms. The app also provides eye care professionals with the ability to search for a needed term, pathology, or assessment.

    Educational materials

    One of the unique features of the Eye Emergency Manual app is a variety of checklists, both for a certain pathology or a patient in general. In the app, users can find a comprehensive list of questions to ask their patients, which is useful both for ophthalmologists and optometrists. Eye Manual also contains pediatric assessment and injured patient assessment.

    What is more, the app developers created a diagnostic tree that is aimed to help users by suggesting diagnoses. After answering a few questions, the app showcases a few diseases and suggests reading about them in the eye atlas.

    OCTaVIA

    mobile ophthalmology app

    One of the main differences between the OCTaVIA mobile ophthalmology app and other apps is the fact that it isn’t free. Some other apps, which I mention in this article, have a paid subscription, but OCTaVIA itself costs $5.99 yearly. However, it is interesting to explore how this price is justified. 

    Eye atlas

    This ophthalmology app contains a constantly updated database of diseases from A to Z. Needless to mention that the application covers only retinal pathologies and provides information about retinal diseases, from Chorioretinal scars to VMT (Vitreo-Macular Traction).

    Educational materials

    One of the advantages of the OCTaVIA mobile ophthalmology app is that for each pathology it provides two views — fundus photo and OCT scan. They may be colored or not, but each fundus photo and OCT scan contains markers, which are explained in the text. What is curious, there are always a few useful links, so users can discover more trustworthy information about the disease.

    Atlas of Ophthalmology Onjoph

    mobile ophthalmology app

    The Atlas of Ophthalmology Onjoph app offers a clinical picture for almost all eye diagnoses. It includes more than 6,000 pathologies, from glaucoma to macular degeneration, and even includes such rare diseases as Stargardt syndrome. The image database is constantly being expanded and updated to include other eye diseases.

    Eye atlas

    Using the search function, eye care specialists can find specific clinical pictures and display them in lists based on diagnoses, ICD-10 code, or keywords. In the Atlas of Ophthalmology Onjoph, users will also find:

    • accompanying diagnosis;
    • code according to ICD-10;
    • brief comment.

    Atlas users can also change the font size, save essential images, or forward images by email.

    Educational materials

    The mobile ophthalmology app has a clear structure for all images. All pathological cases are arranged according to eye regions (conjunctiva, cornea, retina, lens, etc.). Within the eye area, the images are listed according to the type of disease (degeneration, inflammation, tumors, etc.).

    Membership options

    The mobile application also allows its users to save their favorite articles in the Favorites folder, but this feature is paid and has two types of subscription:

    • $3.99 for a Silver plan
    • $29.99 for a Gold plan 

    Other optometrist & ophthalmologist tools worth mentioning

    Ophthalmology Guide

    mobile ophthalmology app

    In case an eye care specialist needs a topic-oriented mobile ophthalmology app, they may check Ophthalmology Guide. Its users are allowed to choose the desired topic and find out the key characteristics of pathologies. In addition, they can also find several fundus photos, scans, and pathology charts.

    Unfortunately, I can’t say that the Ophthalmology Guide app is user-friendly. It contains a few bugs and lacks some additional options, like eye atlases or lectures.

    However, the app is promising thanks to the clear categorization of topics, it can be very convenient for ophthalmologists and optometrists to quickly find specific information about examination and management of the pathology.

    Easy Ophthalmology Atlas

    mobile ophthalmology app

    Easy Ophthalmology Atlas is one of those ophthalmologist and optometrist apps that are also worth mentioning. It is an offline color atlas of the most common eye diseases. The app contains 13 chapters, where users can find clinical features, diagnosis, and treatment management for different pathologies.

    Easy Ophthalmology Atlas lacks quite a lot of features compared to other ophthalmologist tools on the list. 

    However, this mobile ophthalmology app has the potential to replace the heavy paper versions of the ophthalmology guides if the information is updated regularly in it.

    Ophthalmology & Optometry Guide

    mobile ophthalmology app

    Another representative of ophthalmologist and optometrist apps was created to assist students in learning the clinical signs, symptoms, and complications of different pathologies. It provides users with basic knowledge of eye diseases and pathologies, their causes, and treatment.  

    Ophthalmology & Optometry Guide has up to 18 sections, each stands for a specific eye region (conjunctiva, cornea, retina, optic nerve, pupil, etc.). Each section explains the importance of eye region examination and highlights various abnormalities.

    I would recommend this ophthalmology mobile app for beginners or students of the 1st course because it contains a lot of general information that can be useful for those who have just started their careers. However, in the long run, the app lacks media content, real-life examples, and other important features.

    Ophthalmology Atlas

    mobile ophthalmology app

    Ophthalmology Atlas is a database for ophthalmologists and optometrists, showcasing up to 12 areas of eye diseases from A to Z. 

    Here users can find diseases of the cornea, lens, retina, and 9 more. The app is a digital variant of a paper atlas with a bunch of real photos and a lot of complicated cases, which is great for beginners. 

    Clinical Ophthalmology

    mobile ophthalmology app

    The Clinical Ophthalmology mobile app has a very simple interface and a list of 20 pathologies to read about. Although the application has only one feature and lacks media content, the team has provided users with the ability to share content. 

    3D Atlas of Ophthalmology

    mobile ophthalmology app

    The app is a collection of various 3D photos and videos, mostly created by Dr. John Davis. One of the distinctive features of the app is that to watch media content users will need to wear Red-Blue 3D glasses or VR Headset.  

    Will Ophthalmology Mobile Apps Replace Webinars and Conferences?

    According to our research on OCT education, 36% of optometrists and ophthalmologists around the world choose webinars to study OCT interpretation. 36% prefer conferences as the source of new information, 18% choose atlases, and only 11% of eye care specialists trust ophthalmology mobile apps.  

    On the one hand, mobile ophthalmology app cannot replace atlases, webinars, internships, and clinical practice. On the other hand, interactive mobile application contribute to the assimilation of information much better than printed materials and have unlimited data storage capacity. Another of their advantages is that users can learn on the go for little money, while internships and clinical practice takes much time and can be expensive. 

    Summing up, any ophthalmologist and optometrist who has worked at least a little with OCT knows that practical skills are more important than theory. That is why our team believes that ophthalmology mobile apps will inevitably become an additional effective tool for learning OCT interpretation.

  • OCT Interpretation & Eye Examination: How AI can Solve 4 main Problems

    Maria Znamenska
    10 July 2022
    5 min. read

    OCT imaging system is a highly informative non-invasive method of retinal examination, and because of its resolution, it is called histology or microscopy. Usually, thinking of the benefits of OCT eye examination and OCT interpretation, eye care specialists talk about three key points: high scanning speed, non-invasiveness, and the absence of contact.

    Test Artificial Intelliegence for Automated OCT Analysis  Try for Free

    How do eye care specialists learn the interpretation of OCT?

    However, learning OCT interpretation is challenging. It takes time and money to master OCT interpretation skills and become a professional.  Most often, ophthalmologists and optometrists choose one of the following methods of education when it comes to OCT scan interpretation, according to our survey.

    • Webinars. They have become popular with the Covid epidemic. now there is a plenty of various educational webinars where less experienced eye care specialists can obtain useful knowledge.
    • Conferences. Unfortunately, travel restrictions made it impossible to travel much but before the pandemic, eye care specialists could learn by visiting various conferences.
    • Atlases are still quite popular but unfortunately, it is impossible to update information in them often.
    • Mobile apps are a new educational tool that is gaining popularity among eye care specialists.

    OCT interpretation

    Due to the fact that OCT interpretation education requires a lot of resources from eye care specialists, ophthalmologists and optometrists may lack experience they need so much to feel 100% confident with OCT eye examination.

    Poor knowledge of OCT interpretation results in problems

    At Altris Education OCT, we decided to talk to optometrists and ophthalmologists who use our application about the most common problems with OCT eye examinations.  That is what we’ve learned, receiving 1034 answers from eye care specialists from all over the world. There are 4 main problems connected with OCT:

    • No interpretation of OCT

    This problem with OCT interpretation can be hidden, but it turns out that  16, 3 % of eye care specialists avoid offering OCT eye examinations to their clients because they are not sure about their interpretation skills. 

    • Slow OCT

    OCT eye examination takes time and practice to master before an eye care specialist will be able to perform a high-quality OCT examination fast. Some eye care specialists can spend up to 40 minutes on OCT which will result negatively on the quality of the service of the clinic or individual optometry. On average eye care specialists spend 10 minutes on 1 OCT eye examination. 

    AI for OCT

    • Minor, early, rare pathologies missed.

    Another common problem in OCT interpretation is missing minor, early, rare pathologies on OCT scans. It turns out that 20,2% of eye care specialists miss them 1-3 times a week, while 4,4% miss them even more frequently: 3-5 times a week. What is most surprising is how often eye care specialists are not aware of their ignorance at all. 30,5% of ophthalmologists and optometrists admit that they have no idea if they miss any minor, early or rare pathologies at all. 

    If an eye care specialist misses early signs of glaucoma, it can lead to irreversible blindness.

    Why is that so important? Missing pathologies at their early stage can have serious negative consequences for patients. For instance, missing glaucoma, which is irreversible, can lead to blindness. Missing rare and minor pathologies can result in inadequate follow-up and treatment of a patient, which can make the situation worse. Accurate interpretation of OCT scans and diagnosis is the main condition of positive patient outcomes.

    1. Controversial Scans 

    It turned out that a majority of eye care specialists come across controversial scans they don’t know how to interpret. It is difficult to determine the right diagnosis on such scans and additional time is needed to interpret them.

    In the majority of cases ( 99% to be precise) eye care specialists consult their colleagues when they come across a scan they do not know how to interpret. They can ask their colleagues personally, in groups on Social Media or create special chats in messengers.

    How Altris AI can help to solve all these problems?

    With Altris AI, a standalone SaaS for the decision-making support of ophthalmologists and optometrists, all these problems will be solved. Altris AI provides:

    • Fast differentiation between pathological and non-pathological scans
    • Identification of minor, early, and rare pathologies
    • Second opinion when working with OCT scans
    • Confidence when coming across controversial OCT scans

    Our web platform is capable of accurate b-scans severity differentiation. After OCT scans are uploaded inside the system, the AI model assesses them ( up to 512 b-scans) and differentiates between normal scans and scans with moderate and severe pathology.

    The most curious module of our platform is called Classification/Segmentation. Inside this module, an eye care specialist can analyze any OCT scan on the absence/presence of more than 100 retina pathologies and pathological signs. It excludes the possibility of missing some rare pathologies.

    The system is already available for a free trial to anyone who wants to try to solve the main OCT interpretation pain points.

     

  • OCT Examination vs Fundus Photo: Which Method to Choose

    Maria Znamenska
    26 July 2022
    9 min. read

    In the industrialized world, human eye diseases develop quickly and progress rapidly, many of them can lead to blindness. Ophthalmologists and optometrists are well aware that many of these diseases, such as age-related macular degeneration (AMD), diabetic retinopathy, or glaucoma, occur in the retina. That is why the ability to depict the retina, observe its blood circulation and analyze the resulting images is crucial. 

    Test Artificial Intelliegence for Automated OCT Analysis  Try for Free

    Various retinal imaging techniques can improve patient care by increasing the quality of screening and monitoring. Modern technologies, such as fundus photography (FP) and optical coherence tomography imaging (OCT), have a positive effect on the daily practice of ophthalmologists and optometrists, facilitate early diagnosis and allow better management of eye disorders. Currently, special attention is paid to these two methods and their ability to provide a comprehensive description of the morphology and function of the retina.

    At first glance, both methods have great potential for effective screening of retinal abnormalities. However, OCT images of retina provide an improved diagnosis of many diseases, and the role of FP as the gold standard is losing popularity. In this post, we will look at the critical limitations of fundus photography and explore why the OCT imaging system is gaining credibility among ophthalmologists and optometrists worldwide.

    Fundus photography: benefits and limitations

    Being widely available, the fundus imaging system is vital for visualization of retinal and optic nerve conditions. Fundus photography is easy to use and cost-effective, contributing to its rapid spread over the past few years. However, this method also has a few disadvantages which make it less effective than OCT examination. Let’s take a closer look at the benefits and limitations of fundus imaging systems.

    The benefits of the fundus photo

    Fundus photography is a quick and simple non-invasive technique that allows eye care specialists to visualize the retina and provide the accurate diagnosis. FP shows the landmarks of the eye. In addition, fundus photo provides an early and accurate diagnosis, which is highly important for timely treatment and improved therapy. 

    Fundus photography helps ophthalmologists and optometrists not only identify retinal abnormalities and pathologies but also to monitor the progression of eye diseases. In this way, any eye care specialist can develop an effective treatment plan for different people with different eye types.

    The limitations of the fundus photo

    Despite all the benefits of the fundus photo, this technology also has some disadvantages. FP allows eye care specialists to examine the retina by looking at it from above. They may see an uneven retinal surface or curvature. However, FP does not allow observing the microscopic changes inside the retina which correspond to early stages of the disease. It, therefore, can be obtained with OCT image interpretation.

    oct examination

    A key disadvantage of fundus photography compared to optical coherence tomography imaging is its lower resolution. Thus, the pathology size detected in the fundus photography is larger. The FP is unable to detect the invisible pathologies on different retinal layers, which usually present at the stage when the patient does not even have any complaints. In fact, the fundus imaging system sees what the human eye can see. With this technology, an ophthalmologist or optometrist detects only pathologies that are visible to human eyes.

    Main principles of OCT examination

    OCT examination has revolutionized retinal research, allowing doctors to review the pathophysiology of many diseases. But what is the main difference between OCT and fundus photography? FP is the process of photographing the back of the eye using a specialized camera consisting of a microscope attached to a camera with a flash. In contrast, optical coherence tomography imaging estimates the depth at which a particular backscatter occurred by measuring its flight time

    The reflection of light allows determining exactly from what retinal layer the signal is coming. As we know that it takes more time for the light to return from deeper layers. The physical principle of OCT examination is similar to ultrasound. The only difference is that the OCT does not use acoustic waves but near-infrared optical wavelength radiation.

    oct examination

    Modern OCT examination allows doctors to get images with a reasonably high resolution, ranging from 1 to 10 μm. In fact, optical coherence tomography is also called an optical retinal biopsy. The architecture of the retinal structure in the images is very close to the histological structure of the retina. Histologically, the retina consists of 10 layers, but OCT technology allows anyone to assess the retina itself and the structures surrounding it. The modern classification has 18 zones (layers), which can be estimated and described using this technology.

    How OCT examination boosts your working process 

    Modern equipment allows patients to undergo both OCT and fundus photography quite comfortably – without dilation of the pupil and through a non-contact method of research. But optical coherence tomography imaging has many advantages that make this method the most progressive, leaving all competitors behind. 

    OCT imaging system is a highly informative method of retinal examination, and because of its resolution, it is called histology or microscopy. With this technology, ophthalmologists see what could only be seen under a microscope without OCT.

    Advantages of oct examination

    Usually, thinking of the benefits of OCT, eye care specialists  talk about three key points:

    • High scanning speed
    • Non-invasiveness
    • Contactless

    However, experienced ophthalmologists and optometrists know these are not the only advantages. Let’s discuss how OCT image interpretation helps examine the layers of the retina and determine the causes of eye diseases.

    Determining pathologies at early stages

    Many diseases at the early stages are almost invisible to even an experienced optometrist or ophthalmologist. Most retinal abnormalities progress with age and develop slowly and gradually, so diagnosing them is pretty difficult. However, modern OCT image interpretation allows physicians to detect the warning signs of the disease, classify hundreds of pathologies, and re-monitor images to track the progression of pathologies.

    Moreover, OCT image interpretation helps ophthalmologists understand the pathophysiology of retinal diseases, for example, how macular holes arose. This discovery showed doctors that they often misdiagnosed fluid location in the retina. Modern OCT examination help determine the location of abnormal new blood vessels, which is especially important when working with patients suffering from wet AMD.

    oct examination

    Measuring thickness

    OCT imaging allows eye care specialists to measure the retina’s thickness and the magnitude of the pathological process in μm. It is advantageous for the diseases that cause fluid accumulation, such as retinal vein occlusion (RVO) and diabetic macular edema (DME).

    oct examination

    Fundus photography does not provide such an opportunity because the supervision of the dynamics is unavailable in FP. Because OCT imaging allows the retina to be examined in layers, any eye care specialist can detect changes in the structure of the eye that will never be able to be tracked by the FP. 

    In addition, creating a map of the total thickness of the retina or its layers is crucial for monitoring patients with glaucoma, for example. The retinal nerve fiber thickness in such patients becomes thinner as the disease progresses so it is vital to monitor it.

    Determining the severity of eye disease

    Well-made retinal images allow to determine the severity and stage of the disease, compare images after examination with documented results, and track disease progression. Moreover, obtaining clear images of the retina helps different eye care specialists who monitor the same patient to choose the most accurate diagnosis.

    Providing high patient tolerance

    Needless to say that patient cooperation is highly important while performing any type of diagnosis. If a patient moves during the procedure, the quality of the image may deteriorate significantly. However, with modern optical coherence tomography principles, the acquisition time is shorter which results in fewer motion-related artifacts. 

    OCT uses completely safe laser light, avoiding all the side effects or risks. Moreover, with its scanning speed, the process becomes comfortable and effortless both for the ophthalmologist/optometrist and the patient.

    Disadvantages of OCT examination

    Despite the high-quality information provided with optical coherence tomography imaging, the technology also has a few limitations. As OCT uses light waves, some images can contain media opacities. Thus, the OCT scan can be limited by staging a hemorrhage in the vitreous body, a dense cataract, or clouding of the cornea.

    Current use of OCT examinations

    Although standard fundus imaging is widely used, more and more eye care specialists are switching to modern OCT systems that provide more detailed information about various retinal abnormalities.

    Today, the commercially available and clinical standard of choice for most specialists is SD-OCT (spectral-domain OCT) systems, which provide volumetric images of the human retina with a lateral resolution of better than 20 μm. Current SD-OCT devices use retinal images to re-trace the same image area during several subsequent examinations to monitor treatment progress.

    The ophthalmological practice also uses SS-OCT (swept-source OCT) systems, which provide access to a large number of parameters of the eye, which is important for measurements through dense cataracts. SS-OCT supports high image speed and a large scanning depth range compared to SD-OCT. However, the cost of SS-OCT devices is much higher than their counterparts, so these systems have not yet gained widespread clinical implementation. Assuming that the cost of lasers will decrease, it is likely that SS-OCT will eventually also replace SD-OCT in most daily clinical practice.

    In general, the modern OCT devices available today, whether SS-OCT or SD-OCT, are multimodal, which means that ophthalmologists can quickly and easily acquire an incredible amount of information. In addition to image acquisition, modern OCT imagin systems are equipped with special software. It collects retinal images and compares the results to regulatory databases. This allows doctors to make better patient treatment decisions.

    The future of retinal imaging with OCT examinations

    Fundus photo and OCT are pretty difficult to compare because these are completely different technologies. These diagnostic methods carry different information and can sometimes even complement each other. After many years of using the fundus imaging system, this method has been perfected, the quality of cameras has increased, and it has become possible to take pictures without dilating the pupil. 

    For example, FP is a great method for revealing vascular diseases of the eye. However, in most cases, the resolution of OCT is much higher than the resolution of fundus photography. FP will never be able to track invisible changes in the retina structure that OCT can track.

    oct examination

    OCT image interpretation makes it possible to examine 18 zones of the retina, which allows ophthalmologists and optometrists to investigate pathologies in the early stages and detect foci of diseases up to 20 μm. That is why both young specialists and experienced professionals should choose OCT to examine the patient’s retina.

    The future of OCT examination is definitely connected to technologies. 

    For instance, mobile apps for ophthalmologists, such as Altris Education OCT, help eye care specialists learn OCT image interpretation on millions of labeled scans.

    Altris AI web platform supports ophthalmologists and optometrists in decision-making: the system detects 54 pathologies and 49 pathological signs on OCT  providing eye care specialists with a higher level of confidence in diagnostics. 

    The combination of the knowledge of eye care specialists powered by AI technologies will result in higher diagnostic standards for the industry and better patient outcomes. Imagine how many diseases can be prevented if detected at early stages!

Recently Posted

  • machine learning in ophthalmology

    The Application of Machine Learning in Ophthalmology: The View from the Tech Side

    Philip Marchenko
    30.11.2022
    15 min read

    According to the World Health Organization (WHO), artificial intelligence (AI) and machine learning (ML) will improve health outcomes by 2025. There are numerous digital technologies that shape the health of the future, yet AI and ML for medical image analysis look like one of the most promising innovations.

    The healthcare industry produces millions of medical images: MRI, CT, OCT, images from the lab, etc. The right diagnosis depends on the accuracy of the analysis by the specialists. Today AI can back up any medical specialist in medical image analysis: providing confidence and much-needed second opinion.

    Altris AI team decided to improve medical image analysis for just one type of medical image: Optical Coherence Tomography scans of the retina. To do it, the Altris AI team collected thousands of OCT scans and graphically labeled them, defining more than 100 pathologies and pathological signs. 

    machine learning in ophthalmology

    Then all this data was fed inside the AI model, and that is where the magic began… Further, I will tell how exactly we train the AI model of Altris AI so that it can detect more than 100 pathologies with 91% accuracy, but first, let’s discuss first why it is important for the healthcare industry.

    Why is automation with the help of AI important?

    Due to the delicate anatomy of the eye, its treatment carries a high risk of complications. Sometimes these complications can be the result of a medical error by an eye care specialist. But how often?

    According to the Altris team research, 20.2% of eye care practitioners miss minor, early, and rare pathologies on OCT scans 1- 3 times a week, and 4.4% miss them 3-5 times a week. But the worst thing is that 30.5% of ophthalmologists and optometrists are not even sure if they are missing any pathology at all.

    machine learning in ophthalmology

    Some medical errors may be minor, but some may cause significant harm to patients. Such medical errors can lead to medical malpractice lawsuits. That is why most ophthalmic clinics consider implementing AI to double-check the diagnosis of the ophthalmologist. 

    Besides, modern AI tools for medical image analysis have a high level of accuracy and can provide eye care specialists with a second opinion. 

    How to reach a high level of accuracy?

    It is almost always necessary to conduct many experiments to achieve a high level of model accuracy (in the case of Altris AI, it is 91%).  It is often done with the help of a machine learning pipeline.

    machine learning in ophthalmology

    How to reach a high level of accuracy:

    The machine learning pipeline is programmed by a team of engineers to perform certain steps automatically. It systematically trains and evaluates models, monitors experiments, and works with datasets.

      1. ML and Medical teams collect, annotate and preprocess data. It’s crucial to ensure the data quality is at its highest level because the model’s quality heavily depends on it. To do this, the teams developed a process and annotation guideline, which ensures that the number of errors in the annotation is minimized.
      2. ML team chooses the appropriate approach (model) depending on the collected data and the tasks. Each team member is well-versed in the most modern and high-quality approaches that solve emerging tasks.
      3. The selected model is trained on the annotated data.
      4. In the model evaluating and testing stage, we develop tests aimed at helping us understand whether the model is trained properly to perform the needed tasks.
      5. After the ML team is satisfied with the result, we deploy the model, which means the model is ready for production.
      6. While the model is running in production, we monitor its performance to ensure everything goes well.

    This workflow allows engineers to continuously fine-tune existing models alongside constant performance evaluations. The most significant advantage of this process is that it can be automated with the help of available tools. 

    What tasks does machine learning in ophthalmology have?

    Within the Altris AI platform, we solve 2 main tasks: segmentation and classification of OCT scans. 

    Classification task

    Classification is the task of determining which category a particular object belongs to. We assign each pathology to a certain class of pathologies (for example, glaucoma class).

    Segmentation task

    The image segmentation problem can be stated as the division of an image into regions that separate different objects from each other, and from the background.

    Key metrics of Altris ML pipeline

    When discussing classification and segmentation metrics in medical imaging machine learning, it is essential to mention the Confusion matrix (CM). CM is a visualization of our performance, which helps us understand whether the model is performing well in terms of predicted and real data. For a better explanation, let’s take a look at the picture. 

    machine learning in ophthalmology

    Let’s consider 4 possible outcomes from model predictions. Say we need to create a classifier to diagnose or predict if a patient has a disease (positive / 1 or TRUE) or not (negative/ 0 or FALSE). In such a case, the model can predict “yes” or “no”, and we can have an actual “yes” or “no”. Based on this, we can get 4 categories of results:

    • TP — true positive. The patient that actually has a disease has been diagnosed with this disease. A class was predicted to be true, and it is actually true.
    • TN — true negative. The patient is actually healthy and has been diagnosed as healthy. A class was predicted to be false, and it is actually false.
    • FP — false positive (type 1 error). The patient that is actually healthy has been diagnosed as having a disease. A class was predicted to be true, but it is actually false.
    • FN — false negative (type 2 error). The patient that actually has a disease has been diagnosed as healthy. A class was predicted to be false, but it is actually true.

    With the help of the confusion matrix, our ML engineers get specific metrics needed to train our medical imaging machine learning model properly. We discuss each metric in more detail in the following paragraphs.

    Classification metrics

    • Accuracy

    To find out how many of our predictions were correct, we divide the number of correct predictions by the total.

    machine learning in ophthalmology

    While being intuitive, the accuracy metric heavily relies on data specifics. If the dataset is imbalanced (the classes in a dataset are presented unevenly), we won’t get trustful results.

    For example, if we have a training dataset with 98% samples of class A (healthy patients) and only 2% samples of class B (sick patients). The medical imaging machine learning model can easily give you 98% training accuracy by predicting that every patient is healthy, even if they have a disease. Such results may have destructive consequences as people won’t get needed medical treatment.  

    • Precision

    Precision shows what proportion out of all positive predictions was correct.

    machine learning in ophthalmology

    Precision metric helps us in cases when we need to avoid False Negatives but can’t ignore False Positives. A typical example of this is a spam detector model. As engineers, we would be satisfied if the model sent a couple of spam letters to the inbox. However, sending an important non-spam letter to the spam folder (False Positive) is much worse.

    • Sensitivity/Recall

    Recall shows how many of all really sick patients we predicted and diagnosed correctly. It is a proportion of correctly positive predictions out of all positives.

    machine learning in ophthalmology

    In our case, you want to find all sick people, so it would not be so critical if the model diagnoses some healthy people as unhealthy. They would probably be sent to take some extra tests, which is annoying but not critical. But it’s much worse if the model diagnoses sick people as healthy and leaves them without treatment. 

      The sensitivity of Altris AI is 92,51%

    • Specificity

    The specificity shows how many of all healthy patients we predicted correctly. It is the proportion of actual negatives that the medical imaging machine learning model has correctly identified as such out of all negatives.

    machine learning in ophthalmology

    Specificity should be the metric of choice if you must cover all true negatives and you can’t tolerate any false positives as a result. For example, we’re making a fraud detector model in which all people whose credit card has been flagged as fraudulent (positive) will immediately go to jail. You don’t want to put innocent people behind bars, meaning false positives here are unacceptable. 

    The specificity of Altris AI is 99,80%

    Segmentation metrics

    Segmentation also can be thought of as a classification task. For each pixel, we make predictions about whether it is a certain object or not. Therefore, we can talk about Accuracy, Precision, Recall, and Specificity in terms of segmentation. 

    Let’s say we have a Ground Truth (what is really an object) and a Segmented (what the model predicted). The intersection in the picture below is the correct operation of the medical imaging machine learning model. All that is the difference (FN and FP) is the incorrect operation of the model. True negative (TN) is everything the model did not mark in this case.

    machine learning in ophthalmology

    Quite often, even after looking at such metrics, the problem of non-symmetricity remains in the segmentation tasks. For example, if we consider a tiny object, the Accuracy metric doesn’t work. Therefore, segmentation tasks also refer to additional metrics that allow taking into account the size of the object of the overall quality assessment. Let’s look at additional metrics in more detail.

    • Intersection over Union (IoU)/Jaccard

    Intersection over Union is an evaluation metric used to measure segmentation accuracy on a particular image. This metric is considered quite simple — the intersection zone is divided by the union of Ground Truth and Segmented.

    machine learning in ophthalmology

    Sometimes we get such results, like if the object was determined to be very large, but in fact, we see that it is small. Then the metric will be low, and vice versa. If the masks are approximately equal to each other, everything works correctly, and the metric will be high.

    • Dice score/F1

    The dice coefficient is 2 times the area of overlap divided by the total number of pixels in both images.

    machine learning in ophthalmology

    This metric is a slight modification of the previous one. The difference is that, in this case, we take the intersection area twice.

    Calculating scores over dataset

    We calculate the metrics described above for each scan. In order to count them over the entire dataset, we take each picture in this dataset, segment it, calculate the metric, and then take the average value of the metrics on each image.

    What is model validation in ML?

    In addition to evaluating the metrics, we also need to design the model validation procedure suitable for a specific task.

    When we have determined the metric that suits the task of machine learning for medical image analysis, we also need to understand what data to use for calculation. It will be wrong to calculate the metric on the training data because the model has already seen it. This means that we will not check the ability of the model to generalize in any way. Thus, we need a specific test dataset so that we can carry out quality control according to the selected metrics.

    The main tasks of the model validation are:

    • To provide an unbiased estimation of the accuracy that the model can achieve
    • To check whether the model is not overfitted

    Picking the correct model validation process is critical to guarantee the exactness of the validation method. In addition, there is no single suitable validation method for machine learning in ophthalmology — each task requires different validation. Engineers separately examine each task to see if data has leaked from the train dataset to the test dataset because this may lead to an overly optimistic estimate of the metrics.

    For example, we can take OCT images in different resolutions. We may need a higher image resolution for some diseases. If the medical imaging machine learning model overfits at the resolution of this OCT, it will be called a leak because the model should behave the same at any resolution.

    Overfitting and underfitting

    The model also has such an important property as a generalization. If the model did not see some data during training, it should not be difficult for the model to determine which class a certain image belongs to.

    At this stage, engineers may have two problems that they need to solve. The first problem is overfitting. When the model remembers the training data too well, we lose the ability to make correct predictions. The picture below illustrates this problem. The chart in the middle is a good fit when the model is general enough and has a positive trend, and the trend is well-learned. But the chart on the right shows a too-specific model that will not be able to guess the trend.

    machine learning in ophthalmology

    Another problem to solve is underfitting. This problem arises when we have chosen a model that is not complex enough to describe the trend in the data (left chart).

    Bias variance trade-off

    Another important concept we use in machine learning model validation is the bias variance trade-off. We want our models to always make accurate predictions and have no ground truth scattered. As shown at first/second circle.

    However, there are situations when we have a model that predicts something close to the target, but from dataset to dataset, it has a strong scatter. This is showcased in the second circle. 

    In circle three, you can observe a situation where the model has heap predictions on different datasets, but they are inaccurate. This situation usually indicates that we need to almost entirely rebuild the model.

    machine learning in ophthalmology

    Overfitting and bias variance trade-off are very important in working with the model, as they allow us to track errors and select a model that will balance between spread and hitting the target.

    Unbiased estimation

    In addition, within each model, we evaluate a set of parameters. We made a certain estimate (graph on the left), but in real life, the distribution of parameters differs (graph on the right). Thus, seeing that our estimate turned out to be shifted, we find another problem that needs to be solved. Machine learning in ophthalmology needs to make the estimate as unbiased as possible.

    machine learning in ophthalmology

    How do we validate the Altris AI model?

    There are three main steps in choosing a validation strategy for machine learning in ophthalmology:

    • we got familiar with ophthalmology, understood the nature of data, and where the leakages are possible;
    • We estimated the dataset size and target distribution;
    • understood the model’s training complexity (amount of operations/ number of parameters/ time) to pick the validation algorithms.

    After that, we have a reliable strategy for the machine learning model validation. Here are some fundamental concepts we use in the validation of models’ performance.

    Train/test split

    Train/test split is the most simple and basic strategy that we use to evaluate the model quality. This strategy splits the data into train and test and is used on small datasets. For example, we have a dataset of 1000 pictures, 700 of which we leave for training and 300 we take for the test.

    This method is good enough for prototyping. However, we don’t have enough datasets with it to do a simple double-check. This phenomenon is called high sampling bias: this happens when we encounter some kind of systematic error that did not fit into the distribution in the train or test.

    By dividing data into train and test, we are trying to simulate how the model works in the real world. But if we randomly split the data into train and test, our test sample will be far from the real one. This can be corrected by constructing several test samples from the number of data we have and examining the model performance. 

    Train/test/holdout set

    We leave the holdout as the final validation and use the train and test to work with the medical imaging machine learning model. After optimizing our model on the train/test split, we can check if we didn’t overfit it by validating our holdout set.

    machine learning in ophthalmology

    Using a holdout as a final sample helps us look at multiple test data distributions and see how much the models will differ.

    K-fold cross validation

    There is also a more general approach that Altris AI team use for validation — k-fold cross validation. This method divides all of our data equally into train and test.

    machine learning in ophthalmology

    We take the first part of the data and declare it as a test, then the second, and so on. Thus, we can train the model on each such division and see how it performs. We look at the variance and standard deviation of the resulting folds as it will give information about the stability of the model across different data inputs.

    Do we need ML models to perform on par with doctors?

    Here I will try to answer a question that worries many ophthalmologists and optometrists: can machine learning for medical image analysis surpasses an eye care specialist in assessing quality?

    In the diagram below, I have drawn an asymptote called the Best possible accuracy that can be achieved in solving a particular problem. We also have a Human level performance (HLP), which represents how a person can solve this problem. 

    HLP is the benchmark that the ML model strives for. Unlike the Best possible accuracy, for which there is no formula, HLP can be easily calculated. Therefore, we assume that if a model crosses the human quality level, we have already achieved the best possible quality for that model. Accordingly, we can try to approximate the Best possible accuracy with the HLP metric. And depending on this, we understand whether our model performs better or worse.

    machine learning in ophthalmology

    For those tasks that people do better and the ML model does worse, we do the following:

    • collect more data
    • run manual error analysis
    • do better bias/variance analysis

    But when the model crosses the HLP quality level, it is not entirely clear what to do next with the model and how to evaluate it further. So, in reality, we don’t need the model to outperform a human in interpreting images. We simply won’t know how to judge the quality of this model and whether it can be 100% objective and unbiased.

    Avoidable bias

    Let’s say we need to build a classifier for diabetic retinopathy based on OCT scans, and we have a control dataset prepared by people. In the first case, doctors are wrong 5% of the time. At the same time, the model on the train set is wrong in 10% of cases and on the test set — in 13%.

    machine learning in ophthalmology

    The difference between the model’s and the human’s performance is usually taken as the minimum difference between the train/test set and the human. In our case, it is 5% gap (10% – 5%) of avoidable bias. It is called avoidable bias because it can be fixed theoretically. In such a case, we need to take a more complex model and more data to better train the model.

    In the second case, doctors determine the disease with a 9% error. If the model defines a disease with the same rates as the previous example, then the difference between the train/test set and the human will be 1% (10% – 9%), which is much better than avoidable bias

    Looking at these two cases, we must choose a strategy that will lower the variance for the machine learning model so that it works stably on different test sets. Thus, taking into account the avoidable bias and the variance between the samples, we can build a strategy for training the model so that it could potentially outperform the HLP someday. However, do we need it now?

    Understanding HLP

    To better understand the HLP metric, let’s consider the task of determining dry AMD on OCT scans. We have a fixed dataset and 4 train sets, each one determining dry AMD with a specific accuracy:

    • ML engineers – 20%
    • ophthalmologists – 5%
    • 2 ophthalmologists – 3%
    • 2 ophthalmologists and 1 professor of ophthalmology – 2%

    machine learning in ophthalmology

    We take a result of 2% as the best HLP possible. To develop our model, we can choose the performances we strive to get. The 20% error result is irrelevant, so we discard this option. But the level of 1 doctor is enough for model version number 1 model. Thus, we are building a development strategy for model 1.

    Summing up

    Machine learning will revolutionize the eye care industry. It provides confidence and second opinion to eye care specialists in medical image analysis. 

    If you are looking for ways to use machine learning in your eye care practice, feel free to contact us. At Altris AI, we improve the diagnostic process for eye care practitioners by automating the detection of 54 pathological signs and 49 pathologies on OCT images.

  • Altris AI Builds Partnership with Academic Institutions

    Maria Znamenska
    21.11.2022
    2 min read

    At Altris AI we are determined to set higher diagnostic standards for the eye care industry, which is why we are searching for partnerships with academic institutions.

    We would be happy to provide full access to the Altris AI platform in the clinic and education settings with access to the severity differentiation, segmentation and classification, progression, and comparison analysis.

    We can also provide access to the additional feature which will be released over the coming months which includes retinal layer identification and separation, thickness, and volume measurement tools.

    What do you get?

    • The opportunity to test the unique artificial intelligence-powered platform for OCT scans analysis

    • Validation and feedback on the platform with an opportunity for our Head of Product and Chief Medical Officer to have calls with your team to discuss areas about the system development and performance

    • Work with Altris to gain specific scans of pathology that we request so we can train the model – obviously, always GDPR compliant.

    • The opportunity for us to use your logo on our website and promote the partnership with the University on our website, social media, etc.

    • Discuss additional ways that we can work together on future projects.

    Contact us

    Ask us any question
    Required field
    Required field
  • artificial intelligence replace ophthalmologists

    Will Artificial Intelligence Replace Ophthalmologists & Optometrists: Top 5 AI Misconceptions

    Maria Znamenska
    17.11.2022
    8 min read

    Back in 2019, at the RANZCO World Science Congress, Peter van Wijngaarden, Deputy Director of the Center for Eye Research, claimed that the eye sector is one of the leading areas of medicine in terms of artificial intelligence (AI) implementation. According to RANZCO, AI systems are already achieving incredible results and, in some cases, can even rival eye care specialists.

    There are a lot of similar examples when famous professors and specialists in the field of ophthalmology made predictions that AI is rapidly gaining strength in the eyecare industry. This gives rise to many myths and fears around the introduction of AI in clinical practice. More and more eye care professionals have faced the question: will artificial intelligence replace ophthalmologists and optometrists in the near future?

    The increased attention to the issue of optometrists and ophthalmologists replaced by AI was also provoked by a World Economic Forum (WEF) report. According to this report, people can lose 85 million jobs by 2025 due to the shifting division of labor between people and machines.

    In this post, we will discuss the top 5 misconceptions about AI that are most often faced by the owners of ophthalmological clinics and optometry centers in order to dispel them once and for all.

    Do AI algorithms work exactly like a human brain?

    artificial intelligence replace ophthalmologists

    The concept of optometrists and ophthalmologists replaced by robots is gaining popularity. Nowadays, eye care specialists often discuss the potential of AI training in human cognitive skills. It is no longer just about the ability of AI to detect Diabetic retinopathy or interpret OCT scans with greater accuracy. The question is, will AI ever be able to replicate human consciousness? And can AI replicate how the human brain works?

    What do we know about such models in different areas? AI systems are already demonstrating the work of some human cognitive functions. For example, AI models successfully compete with humans in computer games by gradually learning successful strategies. There is also an AI ​​model which creates enjoyable melodic music.

    However, optometrists and ophthalmologists replaced by AI still seem unrealistic. Even with the above examples mimicking some aspects of human behavior, an AI algorithm still needs to learn what empathy is. Artificial intelligence does not understand and cannot make sense of its surrounding, nor can it learn from its surrounding as humans do. The most famous example that confirms this inability of AI is Siri or Alexa. Voice assistants can set up appointments but give strange answers when the conversation goes differently than their scenario.

    While the human brain inspires modern AI techniques such as neural networks (NNs), the structure of NNs architectures is not biologically realistic. 

    First of all, there is a set of qualities that ophthalmologists and optometrists use every day. It is empathy for the patient, as well as creativity, teamwork, and adaptability. These qualities help doctors provide effective care to their patients. It is unlikely that the machines will ever be able to work with children, older adults, or patients with specific disabilities on par with humans. In addition, any patient would like to hear the diagnosis or discuss a treatment plan with a doctor, not a machine. 

    Therefore AI algorithm can’t work like a human brain, and the scenario where artificial intelligence replace ophthalmologists and optometrists will never happen. Nowadays, there are no developments that would make us think that AI image interpretation will ever be able at least to repeat important qualities of eye care specialists.

    Is today’s state of AI dangerous for humans?

    artificial intelligence replace ophthalmologists

    Today, AI algorithms can interpret retinal images and distinguish pathological from non-pathological scans. However, not all attempts at AI implementation have succeeded as well. One of the most popular non-medical examples is Facebook. Some time ago, Facebook tried to identify relevant news for certain groups of users. But the automated process could not detect the difference between real and fake news. Russian hackers managed to trick the system and bypass automatic filters. They posted fake news, forcing the Facebook team to come back to human editors.

    This is just one example of how security lags behind performance when humans rely on AI too much. Artificial intelligence is a great tool, but in most cases, its abilities only give reliable and the most accurate results in collaboration with eye care professionals. Although machines are designed by humans, they often can’t predict human behavior and don’t know how to cope with situations or clinical cases that go beyond the scope of the algorithm.

    Therefore AI is not dangerous for humans when ophthalmologists and optometrists periodically control the work of algorithms and review how the machine works. This is the number-two reason why artificial intelligence replace ophthalmologists and optometrists is unrealistic.

    Will AI ever be 100% objective?

    artificial intelligence replace ophthalmologists

    To honestly answer the questions of will artificial intelligence replace ophthalmologists and optometrists and whether it is 100% objective, you need to understand that an AI system will only be as good as its inputs. By loading unbiased training datasets, engineers can create an AI system that makes unbiased decisions. However, in the real world, AI is unlikely ever to be 100% objective. 

    For example, many well-known companies, such as Amazon or Facebook, still struggle with the gender gap in hiring. Some time ago, Amazon used historical data from the past ten years to train its AI recruiting model. The algorithm was supposed to process data and candidates and free recruiters from the routine viewing of hundreds of CVs. However, soon Amazon team discovered that the data was biased against women. AI algorithm was trained by outdated information when the technology industry used to be dominated by men. Thus, the new recruitment system selected only male candidates. This forced Amazon to abandon the algorithm and re-open many recruiter positions.

    In the field of ophthalmology, AI models can already accurately predict diabetes risk factors or potential vision loss from OCT images. So when will artificial intelligence replace ophthalmologists? In Altris, we are sure the algorithm will never achieve adequate objectivity, as it will always be limited by input data, whether demographics, gender, or age. 

    Now we know that AI can’t be 100% objective. Indeed, ophthalmologists and optometrists can’t match the ability of algorithms to detect pixel-level patterns among the millions of pixels in the OCT scan. However, only the cooperation of eye care specialists and a quality AI model working together will allow for more accurate detection of diseases. The combined efforts of AI management systems and eye care specialists can help achieve the desired 100%.

    Can AI make it without eye care specialists?

    artificial intelligence replace ophthalmologists

    Various articles have speculated on whether artificial intelligence replace ophthalmologists and optometrists, raising concerns about unemployment. However, this never corresponded to the actual state of affairs. Carl Benedikt Frey, an Oxford Martin Citi Fellow at Oxford University, reported that while 47% of jobs are at risk of automation, the risk for doctors is estimated at only 0.4%.

    In addition, in his book “Humans Are Underrated”, Geoff Colvin states that the most valuable skill for ophthalmologists is the ability to sense the thoughts and feelings of patients who are losing sight.

    Many patients complain about the lack of contact with the doctor. They admit that the treatment would be more comfortable if doctors devoted more time to live communication. This mainly applies to children and the elderly, who need a lot of attention from eye care specialists. Empathy and similar human qualities are not only an understanding of the patient’s feelings but also an adequate response to them. Thus, a future in which optometrists and ophthalmologists are replaced by AI seems senseless.

    Professor Tien Yin Wong, medical director of the Singapore National Eye Centre, claimed that AI holds great promise for retinal screening. And while AI for OCT interpretation will radically change clinical practice, the technology’s more significant impact will be to complement and enhance human capabilities rather than replace them. The field of ophthalmology demonstrates that the combined efforts of scientists and machines are more effective than either could achieve individually. 

    Artificial intelligence for OCT interpretation is just a recommendation system for an eye care specialist. Often one pathological sign, for example, Cystoid macular edema (CME), or Intraretinal fluid, can indicate many diseases, like Wet AMD, DR, DME, CRVO, and others. That is why AI is only an assistant to a doctor, especially when it comes to rare pathologies.

    All in all, AI for OCT interpretation is just a tiny part of clinical practice and can never work without humans. In order to detect the pathological signs and diagnose a disease correctly, an eye care specialist must perform different examination methods. Among these exams are visual acuity, intraocular pressure, ophthalmoscopy, and a basic patient examination, which includes anamnesis. Moreover, ophthalmologists and optometrists may also need to perform other visualization methods, like Fundus photography, FFA, or OCTA.

    Will artificial intelligence replace ophthalmologists and optometrists?

    Automation has led to a significant change in many industries, and ophthalmology is no exception. So when will AI replace eye care specialists? The answer is quite simple — AI will never replace them. It will eventually take over routine tasks, allowing the career of ophthalmologists and optometrists to advance in new and exciting directions.

    Automated interpretation of OCT scans will significantly increase the circulation of patients in ophthalmic clinics or optometry centers, which is commercially attractive. Moreover, with increasing life expectancy, and expanding the range and effectiveness of treatment options offered, a collaborative effort between ophthalmologists and AI will improve patient outcomes. This will make ECPs more efficient, freeing up time for human interaction between doctor and patient, which has been a cornerstone of medicine for decades.

    In his concept of the future clinic, Eric Topol describes a system that the Altris AI team is already implementing today. AI labels, annotates and segments images. While ophthalmologists receive information about the structural and functional trends of the patient’s retina to track changes and develop a treatment plan. Altris AI allows ophthalmologists to focus on providing individualized care to each patient.

  • AI medical image analysis

    AI for Reading Centers: How it Boosts Workflow and Efficiency

    Mark Braddon
    05.10.2022
    7 min read

    In recent years reading centers have become an essential resource for facilitating imaging research in many fields, including clinical trials of ophthalmology drugs. And their importance will continue to grow

    Reading centers provide crucial information by evaluating images. That is why for conducting accurate clinical trials, they must hire ophthalmologists of high qualification. Moreover, to ensure consistent analysis, the materials that graders use for the research (be it fundus photographs, fluorescein angiograms, or OCT scans) must also undergo quality control. However, even such measures can’t completely exclude errors or biases.

    Meanwhile, recent developments in the field of AI medical image analysis revolutionized the approach to clinical trials, which makes it possible to boost the workflow of reading centers. AI image analysis software works with thousands of images, efficiently providing the large amount of data needed to analyze the patient’s condition. In addition, evaluating images with AI is faster, cheaper, and more effective

    In this article, we will discuss the top 5 benefits of AI medical image analysis software for reading centers and the way AI improves the image interpretation process.

    Limitations of the manual evaluating procedure

    Although several reading centers have already implemented AI for image analysis in their workflow, most organizations are far from evaluating automation and prefer classic image interpretation methods.

    AI medical image analysis

    In most reading centers, ophthalmologists manually evaluate ocular images for drug safety studies, compile the images, and perform statistical analysis of the data. Research sizes for reading centers can range from 50 images to 3000 or more, and dozen of separate sets of images can be collected per research subject. Therefore reading centers have many obstacles to a quality evaluation process and accurate results.

    • Large amount of images is hard to proceed

    The vast number of images that need to be processed in the short term usually leads to the main problem for reading centers — most hire outsourced ophthalmologists to speed up the image grading and evaluation process. Outsourced specialists have different levels of qualification and different evaluating methods, which may lead to decreased accuracy. In addition, outsourced eye care specialists are not always interested in performing the work at the highest level. 

    • Human resources are expensive

    Another limitation of the standard evaluating procedure is the high сost spent on ophthalmologists. Human resources are usually quite expensive and associated with the risk of staff turnover. 

    • High probability of human bias

    Besides, hours spent in front of a computer screen evaluating thousands of images create a stressful environment for ophthalmologists and cause many errors, affecting the accuracy of the clinical trials. Even the FDA recognizes grader fatigue and its impact on potential errors in image interpretation. 

    • Inaccurate labeling

    In addition, administrative problems also occur quite often. This happens due to deviations from study protocols and incorrect labeling of images, which can compromise the integrity of the analyses.

    Fortunately, the pace of digitalization in reading centers is accelerating. Here is how AI medical image analysis can help reading centers cope with the growing workload. 

    The importance of implementing AI medical image analysis for reading centers

    Usually, AI image analysis is made through a pattern recognition process that involves scanning images for specific pathological signs to interpret the patient’s condition. The AI image analysis software has precise and efficient evaluation protocols that allow the analysis and interpretation of images in terms of a variety of qualitative morphological parameters. For example, when analyzing images of a patient with diabetic retinopathy, the AI models recognize microaneurysms or hemorrhages.

    AI medical image analysis

    AI algorithms allow reading centers to conduct trials of any size and duration, including various treatments for various eye diseases. Moreover, unlike the standard image interpretation process, which requires significant human resources, the introduction of AI for image analysis into the workflow of reading centers has many advantages. 

    • Quality control. Using AI algorithms ensures no errors in OCT scan analysis. AI image analysis software ensures that the desired parameters are classified based on certified imaging protocols.
    • Less money spent. Implementing AI-assisted OCT analysis is less expensive than hiring outsourced ophthalmologists. 
    • Accurate quantification. AI in medical image analysis does not depend on patient characteristics or treatment group assignment knowledge, so the machine provides the most objective and accurate assessment possible.
    • Increased efficiency. Improving the reading centers workflow with AI provides an objective and standardized classification of images. It means that any human bias is excluded, which increases the reputation of clinical research.
    • No time wasted — no more hours spent at a computer screen. Evaluating images with AI medical image analysis provides faster and more sensitive identification of the patient’s condition, which can positively impact decision-making.

    How reading centers will benefit from automated image evaluation

    In short, image evaluation with algorithms is fast, less expensive, and more reproducible. However, many companies that perform clinical trials in cooperation with reading centers are still afraid of implementing AI in medical image analysis and evaluating processes. Modern AI-based image management systems, such as Altris AI, unlike their predecessors, allow reading centers to overcome the challenges of the manual image interpretation process.

    A lot of data available to train an algorithm 

    The more images with various pathological features the algorithm has for training, the more accurately it will detect the diagnosis. Modern AI image analysis software has the ability to obtain thousands of OCT images from different models of devices for comprehensive and correct training of algorithms. Although many medical centers keep their clinical practice confidential, many ophthalmic cases and images with various pathological signs in the public domain allow the training of AI algorithms.  

    For example, the Altris AI medical image analysis software was trained on 5 million unique OCT scans obtained in 11 practicing ophthalmology clinics through the years. Our retina experts took a responsible approach to annotating and labeling images for algorithm training. A thorough error detection and correction procedure gave our algorithm 91% accuracy. 

    Constant quality control

    The responsibilities of the modern algorithms developers include not only the release of the model but also further diagnostics, which allows avoiding the problem of reproducibility. After all, constant quality control is necessary for algorithm development environments. Understanding the importance of quality control, the Altris AI team constantly tests the reproducibility of AI medical image analysis model diagnostics.

    Collection of rare diseases

    According to our research, ​25% of ophthalmologists, on average, miss rare pathologies 3 times a week.​ However, modern AI image analysis software allows overcoming this challenge. For example, Altris AI excludes missing minor, early, rare pathologies. Our team created an algorithm that automates the detection of 54 pathological signs and 49 pathologies.

    High percentage of algorithmic bias is avoided

    Algorithmic bias is one of the biggest challenges in AI. Although algorithms themselves do not have biases, they inherit them from humans. However, today, AI for image analysis has learned how to overcome the lack of interoperability between medical record systems. 

    Although it is impossible to avoid algorithmic bias completely, as it can appear at any stage of the algorithm creation process, from study design and data collection to algorithm development and model selection, modern developers take a direction to fair AI. By using a technical and regulatory framework that provides the diverse data needed to train AI algorithms, the Altris AI team makes modern technologies inclusive and ensures algorithmic bias can be excluded.

    The future of AI medical image analysis in reading centers

    The ultimate goal of the ophthalmic AI system for reading centers is to improve the grading and evaluating process and obtain more accurate research results. However, instead of fully digitalizing image assessment, the ideal approach to analysis is integration — where the benefits of AI algorithms and human skills can be combined.

    Technology will never fully replace humans, but it is already improving their work efficiency. For example, by taking over more routine and monotonous tasks, algorithms allow ophthalmologists to focus on specific eye areas and increase the evaluation speed. AI medical image analysis software can also be effective in determining compliance with the standardization of feature interpretation and determining image quality for requesting more images. 

    There are undoubtedly many challenges to integrating AI for image analysis into the workflow of reading centers. However, modern AI technologies can already overcome almost all of them. Altris AI image interpretation system is changing the future of clinical research by helping to classify images faster and increasing the efficiency, accuracy, and reproducibility of clinical trial data.

  • The use of AI for image analysis

    The Role of AI Image Interpretation for Ocular Pathologies Detection

    Maria Znamenska
    28.09.2022
    20 min read

    The burden of timely diagnostics lies on the shoulders of eye care specialists: ophthalmologists and optometrists worldwide. According to the International Agency for the Prevention of Blindness, over 1 billion people live with preventable blindness because they can’t access the proper diagnostics and treatment. Almost everyone needs access to eye care services during their lifetime. Unfortunately, there are only 331K optometrists worldwide, while 14M optometrists are required to provide effective and adequate eye care services. 

    Test Artificial Intelliegence for Automated OCT Analysis  Try for Free

     

    With the high prevalence of the population that needs eye care services and the lack of specialists, the goal of timely and accurate diagnostics and treatment seems unachievable. 

    However, the empowerment of eye care specialists with Artificial Intelligence (AI) can be a real solution to this problem. As the larger part of the work of eye care specialists relies on retina image assessment and analysis, the support of this process can unburden ophthalmologists and optometrists all over the world. Modern AI image interpretation algorithms, such as Altris AI, can discover patterns among millions of pixels with high speed, accuracy, and zero human errors because of tiredness. 

    In this article, we will talk about the capabilities of AI image interpretation for Optical Coherence Tomography (OCT) in detecting common pathologies, such as AMD or glaucoma, and less prevalent, such as Choroidal Melanoma. Despite the skepticism of the eye care community towards AI, multiple research works mentioned in this article prove the efficiency of AI. Moreover, there are market tools, capable of detecting 49 eye pathologies with 91% accumulative accuracy. Altris, the SaaS created by a team of retina experts based on 5 million OCT scans obtained in 11 clinics, is such a tool. 

     

    AI for Asteroid Hyalosis

    Asteroid hyalosis is a clinical condition in which calcium-lipid complexes are suspended throughout the vitreous collagen fibrils. Although it is a rare disease (​​1.2% prevalence according to the U.S. Beaver Dam Eye Study), it also may lead to unpleasant consequences, such as surface calcifications of intraocular lenses. Today OCT can help with the detection of this degenerative condition. For a higher confidence level, eye care specialists may use Altris AI image interpretation for OCT analysis to detect asteroid hyalosis.

    AI for Central Retinal Artery Occlusion (CRAO)

    Central Retinal Artery Occlusion (CRAO) presents as unilateral, acute, persistent, painless vision loss. It can be bilateral in 2% of the population. The vision loss is abrupt, and the treatment is only effective during the first hours. CRAO resembles a cerebral stroke. Therefore, its treatment should be similar to any acute event treatment: detecting the occlusion site and ensuring it won’t occur again. AI image interpretation models, such as Altris AI, can assist eye care specialists in detecting CRAO today. 

    AI for Central Retinal Vein Occlusion (RVO)

    AI for Central Retinal Vein Occlusion (RVO)

    CRVO is one of the most widespread vascular diseases that affect the population over 45. There are two distinct types of CRVO: perfused (nonischemic) and nonperfused (ischemic). Each of these types has its symptoms and treatment prognosis. For instance, ischemic CRVO leads to sudden visual impairment, while nonischemic CRVO development takes time to develop mildly. The detection of CRVO is now done with the help of OCT predominately, and AI image interpretation systems shows promising results in spotting its symptoms, such as nonperfusion. Altris AI system defines CRVO with 91+% accumulative accuracy in detecting pathological signs that indicate the CRVO.

    AI for Central Serous Chorioretinopathy (CSC)

    Accumulation of fluid under the central retina is called central serous chorioretinopathy. Over time, this disease can lead to the distortion of vision. Fortunately, available AI models for OCT scan analysis show high accuracy in detecting CSC. This and other AI image interpretation models effectively discriminate between acute and chronic CSC, and their performance can be comparable to the performance of ophthalmologists. Altris AI is already helping eye care specialists worldwide to diagnose CSC cases.

    AI for Chorioretinal Scar

    Chorioretinal scars are tiny scars in the back of the eye, the size of which may vary from 0,5mm to 2mm. In most cases, the chorioretinal scar appears as the result of virus infection, such as toxoplasmosis and toxocariasis, or trauma. It usually has no malignant potential. Modern AI image interpretation algorithms allow ophthalmologists and optometrists to diagnose chorioretinal scars more accurately by relying on OCT images.

    AI for Chorioretinitis

    The inflammation of the choroid is called chorioretinitis. Often, the inflammatory process can be caused by congenital viral, bacterial, or protozoan infections. Chorioretinitis is characterized by vitreous haze, fine punctate gray to yellow exudation areas, pigment accumulation along the optic nerve and blood vessels, and flame-shaped hemorrhages with chorioretinal edema. The goal of the eye care specialist is to detect chorioretinitis which can potentially lead to blindness, and to eliminate inflammation. Altris AI image interpretation system can be an excellent decision-making support tool in detecting chorioretinitis.

    AI for Choroidal Melanoma

    Today, choroidal melanoma is the second most common intraocular tumor in the adult population. Patients with choroidal melanoma don’t have distinct symptoms but can have impaired visual acuity, visual field defects (scotomas), metamorphopsia, photopsia, and floaters.

    OCT is a relatively new method for the detection of choroidal melanoma, which is nevertheless gaining popularity. OCT cannot be the only diagnostic method for melanoma detection – FA is also needed for final diagnosis. However, optical shadowing, thinning of overlying choriocapillaris, subretinal fluid, retina local elevation, subretinal lipofuscin deposits, and disrupted photoreceptors can be detected with the help of OCT.

    Such pathological signs will indicate possible choroidal melanoma. Altris AI image interpretation system can assist eye care specialists with detecting pathological b-scans and locating this disease.

    AI for Choroidal Neovascularization (CNV)

    AI for Choroidal Neovascularization (CNV)

    Choroidal neovascularization (CNV) is part of the spectrum of exudative age-related macular degeneration (AMD) and some other conditions. CNV is an abnormal growth of vessels from the choroidal vasculature to the neurosensory retina through Bruch’s membrane.

    Modern OCT systems can detect even a tiny amount of fluid leaking into the retina. Empowered by Altris AI image interpretation algorithm, eye care specialists can spot pathological signs of choroidal neovascularization much faster or detect the pathologies that accompany CNV, resulting in better patient outcomes.

    AI for Choroidal Rupture

    Traumatic choroidal rupture is common after blunt ocular trauma (5 to 10%). It is a defect in the Bruch membrane, the choroid, and the retinal pigment epithelium. The location of the choroidal rupture will define the symptoms: if the fovea and parafoveal retina are included in the rupture area, patients experience impaired vision. In other cases, the rupture can be asymptomatic. OCT is used to diagnose choroidal rupture as it can show the loss of continuity of the RPE layer and the thinning of the choroid. AI image interpretation is exceptionally accurate in layers segmentation and volume/area calculation, so missing the symptom of choroidal rupture with AI is almost impossible.

    AI for Choroidal Nevus

    AI for Choroidal Nevus

    Choroidal nevus is a benign melanocytic tumor of the choroid and is found in 5 to 30% of white people. It can be found accidentally because it is asymptomatic. Artificial intelligence methods are used not only for identifying choroidal nevus but also for early signs of its transformation into malignant melanoma. The earlier the small melanoma is detected, the better the treatment prognosis is for the patient. Altris AI image interpretation system is one of the systems capable of detecting choroidal nevus before its transformation into melanoma. 

    AI for Cone-Rod Dystrophy (CORD)

    CORD is an inherited retinal disease caused by a genetic mutation characterized by cone photoreceptor degeneration. It may be followed by subsequent rod photoreceptor loss. CORD symptoms include loss of central vision, photophobia, and progressive loss of colored vision. OCT diagnostics help to diagnose CORD by pointing at the absent interdigitation zone and progressive disruption and loss of the ellipsoid zone (EZ). Today AI image interpretation is helping to detect Cone-Rod Dystropthy to eye care specialists more confidently, even in controversial cases.

    AI for Cystoid Macular Edema (СME)

    AI for Cystoid Macular Edema (СME)

    Cystoid macular edema (CME) is a painless condition in which cystic swelling or thickening occurs of the central retina (macula) and is usually associated with blurred or distorted vision. CME can be caused by many factors, including diabetic retinopathy and age-related macular degeneration (AMD). OCT diagnostics help to spot СME by detecting retinal thickening with the depiction of the intraretinal cystic areas. CME is not irreversible. Vision loss caused by macular edema can be reversed if detected early. Combining OCT diagnostics with AI image interpretation, eye care specialists can detect CME with higher accuracy at an earlier stage.

    AI for Degenerative Myopia

    AI for Degenerative Myopia

    Degenerative or pathological myopia is the condition during which axial lengthening occurs, especially in the posterior pole. It leads to retina stretching, the sclera’s thinning, choroidal degeneration, and potential loss of vision. AI image interpretation systems have demonstrated excellent results in detecting pathologic myopia and identifying myopia-associated complications on OCT. AI helps ophthalmologists improve the monitoring of pathology treatment and classify different cases of myopia.

    AI for Diabetic Macular Edema

    ​​Diabetic macular edema (DME) is the presence of excess fluid in the extracellular space within the retina in the macular area, typically in the inner nuclear, outer plexiform, Henle’s fiber layer, and subretinal space. DME can develop during any stage of diabetic retinopathy in patients with diabetes.

    Unfortunately, the early symptoms of DME can be unnoticeable or include impaired vision and reading and color perception problems, which some people may ignore. Taking into account its asymptomatic nature, patients with diabetes need regular OCT examinations to determine the presence of DME. OCT has become a golden standard in DME detection within the last few years, and AI can be an excellent decision-making support tool in OCT scans interpretation. According to recent research, AI-powered OCT analysis provides an accurate diagnosis of DME with a cumulative accuracy of over 92%. 

    AI image interpretation systems can unburden ophthalmologists and optometrists who have a lot of patients due to their convenience and can be used in remote regions of the world in the future.

    AI for Diabetic Retinopathy

    AI for Diabetic Retinopathy

    Diabetes can affect the eyes in various ways, most commonly corneal abnormalities, glaucoma, iris neovascularization, cataracts, and neuropathies. However, diabetic retinopathy (DR) is the most common and potentially the most blinding of these complications. Early treatment of both proliferative and non-proliferative DR can improve patient outcomes significantly. OCT is a common diagnostic method for diabetic retinopathy. It relies on the localization of intraretinal and/or subretinal fluid and can help to diagnose diabetic retinopathy through pathological signs detection and layer thickness measurement. 

    AI image interpretation is a step in the future of detection that shows high sensitivity in identifying DR, and studies prove its effectiveness. AI-assisted analysis of OCT scans helps eye care specialists today and will definitely be more widespread tomorrow.

    AI for Dry AMD

    Dry AMD is a more common type of AMD (80% of people have this type), during which patients slowly lose their central vision. It is the aging of the macula and the appearance of deposits called drusen. There is no treatment for Dry AMD yet. However, early detection can help patients to change their lifestyles and slow down the development of this disease. Modern AI solutions make it possible to diagnose Dry AMD faster and develop successful methods of treating the disease. AI image interpretation systems also exclude the possibility of human error.

    AI for Dry AMD – Geographic Atrophy

    Geographic atrophy is an advanced form of the late stage of Dry AMD development. In this condition, retina cells will degenerate and finally die, leading to the patient’s central vision loss.

    AI image interpretation is being widely used to detect Geographic Atrophy with the help of OCT. In this meta research, there are numerous studies that focus on lesion segmentation, detection, and classification of geographic atrophy and even its prediction. They vary in accuracy, but the overall trend of AI for geographic atrophy detection is very positive. The use of artificial intelligence has several advantages, including improved diagnostic accuracy and higher processing speed. 

    AI for ERM or Epiretinal Fibrosis

    AI for ERM or Epiretinal Fibrosis

    Epiretinal fibrosis (epiretinal membrane or macular puckering) is a treatable cause of visual impairment. It is a macula disease caused by fibrous tissue growth on the retina surface. AI image interpretation model for detecting ERM on OCT can outperform non-retinal eye care specialists with a cumulative accuracy of 98+%. For more professional retina experts, AI can be a decision-support tool. Early detection and treatment of this disease are crucial to prevent the ​​growth of fibrous tissue and the worsening of the patient’s condition.

    AI for Epiretinal Hemorrhage

    Epiretinal hemorrhages result from a serious trauma: car or sports accidents, falls, and direct physical impact. Mild hemorrhages unrelated to a serious traumatic event can disappear on their own, but they can be a symptom of a more complex pathology. Epiretinal hemorrhages can be detected with the help of OCT, and AI image interpretation systems can make this process more accurate.

    AI for MTM ( Foveoschisis)

    Myopic foveoschisis or myopic traction maculopathy is the thickening of the retina that reminds schisis in patients with high myopia with posterior staphyloma.

    Untreated foveoschisis often leads to vision loss due to secondary complications, which is why this disease should be detected in time. Today when OCT is becoming more widespread, detection of foveoschisis is more common and accurate. More than that, the studies show that combining the power of AI image interpretation and OCT diagnostics for MTM detection is equal to the junior ophthalmologist’s knowledge. Using AI-powered OCT, it is possible to deal with the shortage of specialists that can guarantee timely diagnostics.

    AI for Full-thickness Macular Hole

    AI for Full-thickness Macular Hole

    A macular hole is a full-thickness defect of the retina involving the foveal region. Patients usually present a reduction of central visual acuity. A complete ophthalmic examination, including OCT, should be performed to diagnose a full-thickness macular hole. So far, the research of AI image interpretation algorithms for a full-thickness macular hole is dedicated to OCT(A), but there are available tools on the market that can help define full-thickness macular hole on OCT scans as well. Altris AI is one of them.

    AI for Hypertensive Retinopathy

    People with high blood pressure, older people, and patients with diabetes often develop hypertensive retinopathy. OCT examination can be used for the detection of hypertensive retinopathy.

    AI-based OCT analysis shows promising results in detecting hypertensive retinopathy by defining retinal vessels and other pathological signs in the retina.

    AI for Intraretinal Hemorrhage

    AI for Intraretinal Hemorrhage

    Among patients with DR, RVO, or ocular ischemic syndrome, there are often those who develop side pathologies. One of these pathologies is intraretinal hemorrhage. AI image interpretation systems help ophthalmologists and optometrists identify intraretinal hemorrhages in the retina.

    AI for Vitreous Hemorrhage

    Vitreous hemorrhage results from bleeding into one of the several potential spaces formed around and within the vitreous body. This condition can follow injuries to the retina and uveal tract and their associated vascular structures. Eye care specialists should perform a complete eye examination, including OCT, slit lamp examination, intraocular pressure measurement, and dilated fundus evaluation. Timely diagnosis and treatment are essential: it can significantly reduce concomitant diseases of intravitreal hemorrhage. AI image interpretation systems can help eye care specialists detect vitreous hemorrhage supporting them in case of controversial OCT scans.

    AI for Lamellar Macular Hole (LMH)

    AI for Lamellar Macular Hole (LMH)

    Lamellar macular hole is one of the types of macular holes known in eye care practice. The problem is that the stage 0 macular hole is a clinically silent finding detected on OCT where a parafoveal posterior hyaloid separation is present and a minimally reflective preretinal band is obliquely inserted at one end of the fovea. Eye care specialists may have problems identifying lamellar macular hole on OCT. That is where AI image interpretation models can come into play.

    AI for Laser-induced Maculopathy

    Since 2014, the number of laser injuries reported worldwide has more than doubled because of the widespread use of laser technologies. Depending on the damage, the patient may have a quick recovery or long-term vision loss with the development of diseases such as photoreceptor’s damage, macular hole, ERM, or others. OCT is one of the methods that help to detect laser-induced maculopathy without human errors and doubts. AI image interpretation models have a reasonable prospect of helping eye care specialists define laser-induced maculopathy based on OCT scans.

    AI for Age-related Macular Degeneration (ARMD)

    Age-related macular degeneration is one of the leading reasons for blindness in people of older age, especially among women and people with obesity. Patients usually present with a gradual, painless vision loss associated with delayed dark adaptation, severe metamorphopsia, and field loss. In other words, in the early stages of AMD, patients may not have any signs or symptoms, so they may not even know they have the disease. Regular OCT screening (among other diagnostic methods) can be a life-saving vest for older people.

    AI image interpretation has shown great promise in detecting AMD, and research papers show that its capabilities are similar to those of ophthalmologists. AI-powered automated tools provide significant benefits for AMD screening and diagnosis.

    AI for Macular Telangiectasia Type 2

    Macular telangiectasia (Mac Tel) results from the capillaries abnormalities of the fovea or perifoveal region related to the retina nuclear layers and ellipsoid zone.

    Macular Telangiectasia Type 2 can have negative consequences and develop into cystic cavitation-like changes in all the layers of the retina or even transform into a full-thickness macular hole. OCT is an effective diagnostic method of macular telangiectasia type 2 as the tomograph can localize foveal pit enlargement. Which is a result of secondary loss of the outer nuclear layer and ellipsoid zone that can progress into large cysts (often called ‘cavitation’) that can encompass all retinal layers.

    Automating the detection of macular telangiectasia type 2 with the help of AI image interpretation systems for OCT scan analysis is already possible thanks to Altris AI.

    AI for Myelinated Retinal Nerve Fiber Layer

    Myelinated nerve fiber layer (MRNF) is a disease that occurs in 1%. It is a benign clinical condition that results from an embryologic developmental anomaly whereby focal areas of the retinal nerve fiber layer fail to lose their myelin sheath.

    OCT is an effective method of MRNF detection with the help of the detection of the RNFL layer. Such tools as Altris AI image interpretation models are even more accurate in retina layers detection and volume measurement thanks to their growing level of accuracy.

    AI for Myopia

    AI for Myopia

    Myopia is not an eye disease. It is an eye-focusing disorder that affects 25% of the world population at a younger age. There are 2 distinct types of myopia: pathological and non-pathological — each of the types has its symptoms and treatment prognosis. The visual function of the patients, as well as the high quality of life, can be preserved if myopia is detected early enough and treated appropriately. Myopia is often diagnosed by ophthalmologists and optometrists with the help of OCT, thanks to its fine cross-sectional imagery of retinal structures. Unlike biomicroscopy, angiography, or ultrasonography, OCT can reveal undetectable retinal changes in asymptomatic patients with myopia.

    Current AI image interpretation models show great promise in detecting myopia on OCT scans, and their results can be compared to the results of junior retina specialists. Altris AI is an accurate AI tool for myopia.

    AI for Pigment Epithelium Detachment

    Retinal pigment epithelial detachment (PED) is often observed in Wet AMD and other conditions. It is determined as a separation of the RPE layer from the inner collagenous layer of Bruch’s membrane. With its capability to visualize retinal layers, OCT helps eye care specialists with timely PED diagnostics. Powered with AI image interpretation systems, OCT diagnostics can promise zero human errors and exceptional accuracy.

    AI for Polypoidal Choroidal Vasculopathy (PCV)

    Polypoid Choroidal Vasculopathy is a disease of the choroidal vasculature. Serosanguineous detachments of the pigmented epithelium and exudative changes that can commonly lead to subretinal fibrosis are the main OCT signs of PCV. AI image interpretation systems show great potential in establishing a difference in diagnostics between PCV and AMD.

    AI for Preretinal Hemorrhage

    AI for Preretinal Hemorrhage

    Preretinal hemorrhage is a complication of many pathologies, such as leukemia or ocular/head trauma. Missing preretinal hemorrhage means putting a patient at risk. Preretinal hemorrhage can be a presenting sign of some systemic diseases. In any case, OCT diagnostics are performed to determine preretinal hemorrhage and its real reason.

    AI for Pseudohole

    Sometimes the pulling or wrinkling of the epiretinal membrane (ERM) can result in a gap called a pseudohole. A pseudohole can look like a macular hole; sometimes, it can turn into one, so it is essential to distinguish between these two phenomena. Optical coherence tomography can accurately determine a pseudohole revealing an epiretinal membrane with contraction of the retina or suppression of retinal layers. Combined with AI image interpretation, OCT diagnosis can guarantee higher accuracy in pseudohole detection.

    AI for Retinal Angiomatous Proliferation (RAP)

    RAP is a subtype of AMD, which is neovascularization that starts at the retina and progresses posteriorly into subretinal space. There are 3 stages of RAP: intraretinal neovascularization (IRN), subretinal neovascularization (SRN), and choroidal neovascularization (CNV). OCT is effective for detecting IRN only since changes beneath the pigment epithelium are challenging to assess. AI image interpretation models effectively differentiate between RAP and polypoidal choroidal vasculopathy (PCV), comparable to the performance of eight ophthalmologists. AI cannot substitute eye care specialists but can be an excellent decision-making support tool.

    AI for Retinal Detachment

    Retinal detachment is a serious eye condition that happens when the retina pulls away from the tissue around it. It can be a result of trauma or another disease. OCT has become a new standard for detecting early retinal detachment and defining the best time for surgical operation, for example. OCT powered with AI image interpretation systems can give eye care specialists the confidence they need to determine the degree of detachment and make the correct prognosis.

    AI for Retinitis Pigmentosa

    RP is a hereditary diffuse pigment retinal dystrophy characterized by the absence of inflammation, progressive field loss, and abnormal ERG. OCT diagnostics allows assessing morphological abnormalities in RP, providing insights into the pathology of RP and helping to make a good prognosis. AI image interpretation applied for the OCT analysis shows promising results in Inherited Retinal Diseases detection and future management.

    AI for Retinoschisis

    AI for Retinoschisis

    Retinoschisis is an eye condition characterized by a peripheral splitting of retinal layers. OCT is an effective method of retinoschisis diagnostic. The application of AI image interpretation tools for OCT analysis for identifying retinoschisis (among other myopia conditions) is comparable to the performance of experienced ophthalmologists.

    AI for Retinal Pigment Epithelial (RPE) Tears (Rupture)

    RPE rupture or RPE tears is the condition when this retinal layer acutely tears from itself and retracts in an area of a retina, usually overlying a pigment epithelial detachment (PED). OCT is an effective method of diagnostics of RPE tears. OCT scans will show a discontinuity of the hyperreflective RPE band, with a free edge of RPE usually wavy and scrolled up overlying the PED, contracted back towards the CNVM.

    AI image interpretation systems can provide eye care specialists with confidence when detecting RPE tears. Systems such as Altris AI can distinguish between retinal layers with exceptional accuracy, exceeding the accuracy of eye care specialists.

    AI for Solar Retinopathy (Maculopathy)

    Solar retinopathy is photochemical toxicity and the consequent injury to retinal tissues located in the fovea in most cases. OCT helps to diagnose solar retinopathy by indicating changes and focal disruption at the level of the subfoveal RPE and outer retinal bands. The overall retinal architecture remains intact. AI image interpretation models can confidently assist eye care specialists in detecting solar retinopathy, even when they are in doubt.

    AI for Subhyaloid Hemorrhage

    Subhyaloid hemorrhage is diagnosed when the vitreous is detached from the retina because of blood accumulation. This type of hemorrhage is rare and is different from intraretinal hemorrhage caused by trauma or diabetes. OCT helps to detect subhyaloid hemorrhage. For eye care specialists who don’t have experience in detecting subhyaloid hemorrhage, the AI image interpretation model can become a great support tool.

    AI for Subretinal Fibrosis

    Subretinal fibrosis appears due to wound healing reaction to the choroidal neovascularization in nAMD or other conditions. Early diagnostics of subretinal fibrosis are critical because a neovascular lesion’s transformation into a fibrotic lesion can be very rapid. OCT is regarded as the most accurate method of diagnostics today.

    AI image interpretation systems can help eye care specialists who use OCT with early diagnostics of subretinal fibrosis and improve patient outcomes.

    AI for Subretinal Hemorrhage

    AI for Subretinal Hemorrhage

    Subretinal hemorrhages are a complication of various diseases which arise from the choroidal or retinal circulation. It is most often caused by AMD, trauma, and retinal arterial macroaneurysm. OCT will be an effective tool for determining the level at which subretinal hemorrhage occurred. Powered with AI image interpretation models, OCT can become the decision-making support tool eye care specialists need for subretinal hemorrhage identification.

    AI for Sub-RPE (Retinal Pigment Epithelial) Hemorrhage

    Sub-RPE (retinal pigment epithelium) hemorrhage is located between the RPE and Bruch’s membrane. OCT is an essential tool for validating the hemorrhage’s diagnosis and localization. AI image interpretation tools, such as Altris AI, will ensure that Sub-RPE hemorrhage is not missed.

    AI for Tapetoretinal degeneration or dystrophy

    Tapetoretinal dystrophy or tapetoretinal degeneration (TD) is exogenous destruction of the retina caused by a genetic mutation. Eye care specialists might easily miss such rare conditions as tapetoretinal degeneration. It is often advisable to have AI image interpretation systems as a decision-making support tool not to miss TD or other uncommon diseases.

    AI for Vitelliform Dystrophy

    It is autosomal dominant degenerative maculopathy wherein a mutation in the bestrophin gene leads to lipofuscin accumulation in RPE cells manifested in a yellow spot. Detecting vitelliform dystrophy is critical at the early stages as it can lead to vision loss. OCT provides essential information on the lesion’s morphology, location, and dynamics. Empowered with AI image interpretation tools, such as Altris AI, eye care specialists won’t miss such a rare disease as vitelliform dystrophy at the early stage.

    AI for Vitreomacular Traction Syndrome

    Vitreomacular traction syndrome is a pathological condition characterized by a posterior vitreous detachment that leads to blurred vision or serious vision impairment. OCT is an essential method of diagnostics of vitreomacular traction syndrome as it can show the amount of involvement and tension on the macula caused by VMT. Combined with AI image interpretation tools, OCT analysis can give incredible results.

    AI for Wet AMD

    AI for Wet AMD

    Wet AMD is the most widespread disease among the elderly population in developing countries. It is a disease characterized by abnormal blood vessel growth under the retina. Understanding that this disease can lead to rapid and severe vision loss, its early detection and treatment are very important. OCT is a golden standard for the diagnostics of wet AMD as it shows fluid or blood underneath the retina without dye, among other pathological signs.

    Today AI shows promising results in predicting the development of wet AMD based on OCT images. For instance, the DARC algorithm designed for detecting apoptosing retinal cells could predict new wet-AMD activity. Another effective AI image interpretation algorithm determines the location and volumetric information of macular fluid within different tissue compartments in wet AMD, providing eye care specialists with the ability to predict visual acuity changes.

    AI for X-linked Juvenile Retinoschisis (XLRS)

    XLRS is a rare congenital retina disease caused by mutations in the RS1 gene, which encodes retinoschisin, a protein involved in intercellular adhesion and likely retinal cellular organization. The disease usually affects younger males in their teenage years who complain about blurred vision. OCT is used to detect schisis in the superficial neural retina and thinning of the retina. Despite the lack of research articles on AI in OCT diagnostics of XLRS, there are AI image interpretation tools that already cope with this task effectively.

    Final Words

    Artificial intelligence can identify, localize, and quantify pathological signs in almost every disease of the macula and retina. That is how AI image interpretation systems can provide decision-making support with the pathologies at their early stages or rare pathologies. AI can help to detect many pathologies that are invisible to the human eye because of their size or that are at their early stage. 

    The overall potential of artificial intelligence for ophthalmologists and optometrists is enormous and includes pathological scan selection and scan analysis with the probability of existing pathologies and pathological signs. One trial is worth a thousand words in the case of AI tools for ophthalmologists and optometrists.

  • oct scan interpretation

    Types of Optometry Eye Examination Techniques & the Role of OCT

    Mark Braddon
    14.09.2022
    7 min. read

    Various optometry eye examination techniques have always played a crucial role in diagnosing many eye diseases and promptly referring to a retinal expert. According to Essilor International research, poor vision is the most common disability in the world today. The good news is that 90% of vision loss cases are treatable or preventable if discovered in their early stages.

    However, by performing only traditional optometric eye examination techniques, such as anterior and posterior segment examinations, optometrists may miss the complete picture of a patient’s eyes. That is why optometry specialists are embracing a new technique: optical coherence tomography (OCT) examination. 

    Test Artificial Intelliegence for Automated OCT Analysis  Try for Free

    OCT imaging helps go beyond standard eye examination procedure by better visualizing the eye’s structures and providing an additional quantitative assessment.

    In this article, I will discuss the most important optometric examination and emphasize the role of OCT scan interpretation in optometry.

    Types of optometry eye examination techniques

    When performing a full optometric examination, the optometrist should not only assess the visual acuity with an eye chart but also check their eye health. The types of eye examination tools are now very diverse and depend on the application field and the qualification level. Nowadays, there are a few eye examination techniques, although they may vary from country to country, that help diagnose a patient more accurately and improve follow-up care.

    Ophthalmoscope examination

    eye examination techniques

    Ophthalmoscopy plays a crucial role in detecting the conditions of the retina, blood vessels, and optic disc. This is a basic eye examination procedure that optometrists usually perform to evaluate many diseases, such as diabetic retinopathy or retinal vein occlusion. 

    During the direct ophthalmoscopy, the optometrist shines a light into the patient’s eyes to see the inside. Binocular indirect ophthalmoscopy also involves shining a light into the patient’s eyes, however, it allows eye care specialists to take a better look at the retina and its parts that are difficult to see with other eye examination techniques. The indirect ophthalmoscopy is usually combined with pupil dilation and another optometry practice called scleral depression.

    Slit lamp examination

    eye examination techniques

    A slit lamp consists of a microscope, light source, and frame on which a patient lies their head. This regular eye examination procedure lets an optometrist focus on the eye by working with the light: expand or narrow it, increase brightness, and filter with colors. Sometimes the procedure also includes putting a few dye drops in a patient’s eye to examine some of its parts.

    Slit lamp examination is pain-free and allows an optometrist to view the sclera, iris, or cornea to detect diseases related to allergies, autoimmune disorders, gout, or even melanoma. Such eye examination procedure also allows to view the retina of the eye to detect the pathological signs of diabetes. Optometrists usually use a slit lamp along with an ophthalmoscope examination.

    Refraction test

    eye examination techniques

    One more type of eye examination techniques is a refraction test, usually performed to detect if a patient needs glasses or contact lenses. This test made with a phoropter is quick and painless. During the optometric examination, the optometrist adjusts the power of the lenses by moving or turning them back and forth until a patient can clearly see the letters on the chart.

    An optimal value of 20/20 is considered ideal vision, while a deviation means a refractive error. This may indicate that when light passes through the lens of the patient’s eye, it is not refracted properly. An optometrist can detect astigmatism, myopia, presbyopia, and a refractive eye problem during a refraction test. This, in turn, helps detect macular degeneration, retinal vein occlusion, retinitis pigmentosa, and retinal detachment.

    • Cycloplegic refraction

    Sometimes the optometrist may decide that the normal refraction is insufficient or inaccurate due to error. During refraction, the patient may unconsciously focus, affecting the test result and showing nearsightedness or farsightedness.

    Then the optometrist performs cycloplegic refraction using cycloplegic eye drops. This eye examination procedure paralyzes the muscles that focus the eye to determine the refractive error. Сycloplegic refraction exam is especially useful for children, patients with pre-presbyopia, and LASIK patients.

    • Autorefraction

    Autorefraction is an eye examination procedure performed using a special autorefractor device, also called an optometer. This exam automates the estimation of refraction and determines its error. Usually, the indications for the procedure are myopia, farsightedness, astigmatism, presbyopia, and prescription of glasses and contact lenses.

    Retinoscopy

    eye examination techniques

    One more optometric examination usually performed to detect farsighted, nearsighted, or astigmatism, and the need for glasses is retinoscopy. This procedure is pain-free and quick. Using a retinoscope, the optometrist projects a beam of light into the patient’s eye. This light moves along a horizontal and vertical trajectory, reflecting off the back of the eye. The eye care practitioner observes the movement of light with the help of lenses they place in front of the eye. Then the optometrist changes the lens’s power and tracks the reflection’s direction and pattern. This test is performed to find a possible anomaly.

    Role of OCT scan interpretation in optometric examination

    The types of optometry eye examination techniques described above are fundamental for any diagnosis. However, adopting modern OCT scan interpretation systems already complements clinical practice perfectly and has the prospect of widespread distribution among optometrists worldwide. 

    Knowing that the prevalence of some eye conditions, such as Myopia or Dry AMD, has increased with the pandemic, specialists need to implement modern methods and eye examination techniques in their clinical practice. Current optical coherence tomography devices allow optometrists to perform consistent analysis and furthermore have special software and a database for storing patient information. Compared to other retinal examination methods, such as fundus photography, OCT scan interpretation enhances patient care by improving the quality of diagnosis.

    High-quality information provided

    Modern OCT diagnostics allow optometrists to quickly obtain a huge amount of information about the patient’s eye. Built-in software collects images and compares results to normative databases. This allows optometrists to easily track patient progress or regression and generate reports that ophthalmologists or surgeons may need for follow-up treatment.

    For example, suppose a patient has a disorder with the optic nerve, macula, or vascular system. In that case, the optometrist can send data to the ophthalmologist promptly, highlight important aspects of the patient’s condition, and provide abnormal OCT scan results for additional clarity. 

    No missed pathologies

    OCT provides higher diagnostic standards, ensuring fewer pathologies or pathological signs are missed. OCT scan interpretation helps detect early vision-threatening eye conditions. For example, the system can detect AMD in the early stages, which is crucial for preventing vision loss due to subretinal fibrosis. With OCT, the thickness of the retina over the macula and posterior pole can be analyzed to detect retinal edema or atrophy. Optometrists can also confirm diabetic macular edema (DME) and decide on further treatment based on the results of its examination. In addition, OCT perfectly visualizes the retinal pigment epithelium (RPE) and choroid.

    More patients served with comfort

    By better visualization of the eye structures, optometrists provide each patient with an individual approach. This level of service ensures comfort for patients and trust for a specialist. OCT allows optometrists to avoid routine work and devote more time and energy to patients. More importantly, the OCT scan interpretation helps establish contact, allowing patients to understand the examination and treatment plan.

    Impact of AI on OCT eye examination techniques

    OCT scanning allows optometrists to accumulate large amounts of patient data. However, a large amount of information can be difficult and time-consuming to process, even for experienced specialists. The collaboration of OCT and artificial intelligence (AI) gives optometrists a unique opportunity to analyze a large amount of data and make better clinical decisions. Here are 4 key benefits of AI which completely transform the OCT scan interpretation process for optometrists:

    • Gaining confidence. 16.3% of interviewed eye care practitioners still avoid using OCT in their daily practice because of the lack of confidence in their interpretation skills. However, with AI, this problem will be solved.
    • Fast examination. Implementing AI-powered management systems in daily clinical practice reduces the time optometrists have to spend on non-pathological scans.
    • Clear diagnosis. 59% of specialists acknowledge that they have to interpret controversial scans around 1-3 times a week. AI helps optometrists with controversial and abnormal OCT scans, so they don’t need to guess the diagnosis.
    • High diagnostic standards. 30,5% of interviewed ECPs admit they are unsure how often they miss pathologies. When working with OCT, AI systems ensure no minor, early, rare pathologies are missed.

    OCT scanning allows specialists to easily, quickly, and safely obtain many images, producing a lot of data. As AI aims to work with large volumes of data, more and more AI models are being created to help optometrists.

    eye examination techniques

    Altris AI has developed an artificial intelligence platform to assist ECPs during their optometric examination and already plays a significant role in diagnosing and treating eye diseases using OCT. We have trained an AI algorithm on 5 million OCT scans collected in 11 ophthalmic clinics with a 91% accuracy.

    Future of OCT scan interpretation in optometric examination

    The integration of OCT into the clinical practice of optometrists is beneficial and shows great promise. However, to gain the most accurate diagnosis, the interpretation of scans should be carried out in cooperation with other optometry eye examination tools. Optical coherence tomography implemented with other eye examination techniques, including gonioscopy or slit lamp biomicroscopy, boosts diagnostic performance and provides valuable data.

    OCT scan interpretation in optometry practice is becoming routine for providing improved examination and patient care. This technology can also improve the confidence of eye care specialists. Detecting many pathologies using optical coherence tomography has an immediate practical benefit. Due to its high resolution, it defines and identifies early pathological signs before patients even notice any symptoms.