Lecture: Gonioscopy and Angle Evaluation: Fundamentals to Advanced Techniques & Technologies

Hello, and good day, everyone. I’m Dr. Benjamin Xu, Chief of the Glaucoma Service and Director of Data Science and Artificial Intelligence here at the University of Southern California Roski Eye Institute.

It’s my pleasure today to give this talk on gonioscopy and angle evaluation, Fundamentals to Advanced Techniques and Technologies.

I’d like to thank Orbis and Dr. Cherwek for inviting me to give this talk today.

Before I begin, I’d like to acknowledge my funding sources, including the National Eye Institute and various foundations.

I do receive research support from Heidelberg Engineering, which manufactures anterior segment OCT devices, and I will be talking about the technology today.

This is the outline for my talk. I’ll first start by presenting a little bit about gonioscopy, how it should be performed, and proper technique.

Next, I’ll talk about primary angle closure, the spectrum of disease that is visually disabling and a public health issue worldwide.

Third, I will talk about anterior segment OCT, which is an exciting technology that promises to help us deliver more precise care to our angle closure patients.

I’ll talk about how to interpret these types of images and incorporate them into daily clinical practice.

And finally, we’ll talk about artificial intelligence, which is a transformative technology that promises to change the way that we practice ophthalmology and medicine as a whole.

And I’ll end by talking about the future of anterior segment OCT technology, AI, and angle closure care.

Let me first start by talking a little bit about gonioscopy.

When you perform gonioscopy, you’re enjoying a small slice of ophthalmology history.

This exam technique was pioneered by Maximilian Salzmann more than a century ago. And as described by Professor Lee Alward in his A History of Gonioscopy, this technique has been invaluable for the glaucoma evaluation for over a century.

And really, the technique has not changed much since it was first introduced.

It is still performed at the slit lamp and is a crucial part of the glaucoma evaluation.

The key determination when performing gonioscopy is to identify whether an angle is open or closed.

This is the first fork in the decision tree when we are evaluating our glaucoma patients and coming up with a care plan for them.

The importance of gonioscopy has been emphasized by the American Academy of Ophthalmology in their Preferred Practice Pattern guidelines.

The AAO recommends that gonioscopy should be performed in all patients who are suspected of glaucoma, suspected of angle closure, or who will be receiving incisional glaucoma surgery.

Therefore, it is crucial that we understand the fundamentals of gonioscopy and how to perform this examination technique properly.

And I’m going to be highlighting five different components of the gonioscopic exam: the tear meniscus, lighting, indentation, tilting, and grading.

The first thing to recognize is that the tear meniscus is essential for visualizing the angle.

Due to an optical principle called total internal reflection, if we don’t have a smooth tear meniscus, then we can’t visualize the anatomical structures inside the eye. Therefore, it’s important to squeeze out any air bubbles gently prior to initiating the gonioscopic exam.

Next, it’s also important to recognize that the lighting conditions determine the width of the angle and that the angle widens in the light.

This is a video that demonstrates this principle.

You can see here that in the light, the trabecular meshwork is visible; when the light is dimmed by narrowing the slit beam, we can see that the configuration of the angle changes. And then, once the light is turned back on, the iris is drawn away from the angle recess as the pupil constricts. So it’s very important to recognize that gonioscopy should be performed ideally in a dark environment in which the angle is maximally narrowed.

Next, indentation helps differentiate between appositional and synechial closure.

Indentation is a very important component of dynamic gonioscopy.

And the principle here is that by applying a force on the corneal surface, the aqueous humor in the anterior chamber, which is a non-compressible fluid, is then forced into the angle recess, opening the angle recess, as long as there is not the presence of peripheral anterior synechiae. And so this is demonstrated in this video here in which you can see that, when there’s indentation, you can see the pigmented trabecular meshwork. And then, when the indentation is lifted, you cannot see the pigmented trabecular meshwork anymore.

And so I’m going to play this video one more time just so you can appreciate this effect.

So here, no angle structures are visible. And then on indentation, you can see the pigmented trabecular meshwork. And so this is an important step in differentiating between synechial closure, in which indentation would not widen the angle, and appositional closure, which can be seen here.

Next, tilting helps overcome a high lens vault.

The position of the lens determines how easy it is to visualize the structures in the angle recess.

For example, here is an eye with a high lens vault. And when you’re performing gonioscopy, it might feel like you’re standing on the wrong side of the hill and that you want to get over this hill in order to look into the angle recess. The proper tilting technique is that you want to tilt the gonioscopy lens toward the angle quadrant that you are trying to visualize, in essence, peering over the hill.

This is another way of conceptualizing it in which you are starting first on the wrong side of the hill, and then as you walk up to the top of the hill and peer down into the angle recess by tilting toward the quadrant that you want to visualize, then you can see the structures of the angle at that point.

Finally, it’s important to have a grading system to describe your gonioscopic findings.

Clearly, the structures that we’re looking for are typically the trabecular meshwork or the scleral spur in order to identify the degree of angle narrowing, but this needs to be described in a rigorous fashion. And so there are different grading systems that exist to describe our findings.

One commonly used system is the modified Shaffer grading system, which simply describes the anatomical structure that can be visualized. And this is the most anterior structure. And so grade four is typically described as all the way up to the ciliary body.

And then grade three is to the scleral spur. Grade two is the pigmented trabecular meshwork, and grade one is to the non-pigmented trabecular meshwork, and grade zero is when no structures are visible. And so this becomes a very convenient way of describing what you see on exam.

There are other grading systems, such as the Spaeth grading system, that are more complicated but also more precise in terms of defining not only what is visible but also the exact configuration of the structures in the angle. Of course, now we have OCT imaging and ultrasound biomicroscopy, which can accomplish many of these same goals.

And so, these types of grading systems perhaps are more complicated than what are typically used in the clinical setting.

It’s important to recognize that gonioscopy reveals many abnormalities, some of which cannot be seen using anterior segment OCT.

One key finding is the presence of peripheral anterior synechiae or PAS, which can be easily seen on gonioscopy, sometimes requiring indentation but can be hard to see on AS-OCT imaging. So here on the left, you can see pinpoint PAS, and on the right, you can see slightly broader PAS.

And these are typically quite easy to visualize on gonioscopy.

Other abnormalities that can be easily visualized would be dense pigmentation that might be suggestive of pigment dispersion or pseudoexfoliation.

And on the right, you can see an area of PAS, neovascularization of the angle, and these could be missed using imaging modalities.

And, of course, it’s very important to perform gonioscopy to detect pigmented lesions in the iris that might be suggestive of uveal melanoma.

These can often be missed on slit lamp exam and may not be so visible on anterior segment imaging as well, therefore highlighting another important aspect of gonioscopy.

Next, I’d like to talk about primary angle closure, which is a primary reason that we perform gonioscopy and is an important distinction when we are diagnosing patients with glaucoma.

Primary angle closure glaucoma or PACG is one of the most common causes of irreversible vision loss worldwide.

The condition currently affects around 24 million people, and this condition is endemic in Asia where approximately three-quarters of PACG cases can be found.

The prevalence of PACG is rapidly rising due to the aging of the world’s population, and it’s expected that 34 million people will have this condition by the year 2040.

And this rise in PACG prevalence is concerning as PACG is more visually blinding than POAG, conferring a 2.4 times higher odds of blindness compared to POAG. And this is related to the high intraocular pressures that are typically associated with angle closure.

In angle closure, the primary problem is that the peripheral iris is coming forward and contacting the trabecular meshwork.

This impedes the normal outflow of aqueous from the eye. And when the aqueous is trapped in the eye, then this can cause the intraocular pressure to rise, thereby leading to damage of the optic nerve.

And this is angle closure glaucoma.

Angle closure detected on gonioscopy forms the foundation for current definitions of the spectrum of primary angle closure or SPAC.

Primary angle closure suspect or PACS is the mildest form in which there are two or more quadrants of angle closure on gonioscopy.

This is generally thought to be a low-risk state for PACG.

Next is primary angle closure in which there is gonioscopic angle closure in two or more quadrants, in addition to the presence of PAS or elevated intraocular pressure, typically defined as over 21 millimeters of mercury. And this is thought to be a high-risk state for PACG.

Finally, primary angle closure glaucoma or PACG is diagnosed when there is the presence of optic nerve damage, with or without visual field defects in addition to the gonioscopic findings of angle closure in which pigmented trabecular meshwork cannot be visualized.

And so PAC and PACG together form primary angle closure disease, or PACD, which is the more severe part of the spectrum of primary angle closure.

Primary angle closure glaucoma is commonly associated with blindness. This figure shows the prevalence of blindness associated with PACG in Asian regions, based on data from population-based epidemiological studies.

We can see here that the prevalence of blindness exceeds 25% in many countries.

The exception to this rule is in countries that have better access to eye care services and cataract surgery, such as in Japan and Singapore, where we see a lower prevalence of blindness among PACG patients.

Until recently, not much was known about the burden of PACG here in the United States.

My lab conducted a study using data from the IRIS Registry, which is an electronic healthcare records database that is administered by the American Academy of Ophthalmology. And this database contains data on over 70 million Americans who are receiving eye care. And within the IRIS Registry, we identified just over 49,000 patients who were newly diagnosed with PACG between 2015 to 2019.

When we evaluated these cases, we found that 12.4%, or approximately one out of eight of these patients, was blind in at least one eye at the time of first PACG diagnosis. And so this is despite very high healthcare expenditures here in the United States.

We also found that there were disparities in the severity, with a higher risk of blindness among Black and Hispanic Americans, thereby highlighting the disproportionate effect of this disease on some racial populations here in the US.

So these data highlight an important issue in the field of angle closure, which is that PACG often remains undiagnosed and untreated until after severe vision loss has occurred.

And this also highlights the limitations of gonioscopy. Despite being the longstanding clinical standard for evaluating the angle and detecting angle closure, gonioscopy does have some limitations that we need to keep in mind. First of all, gonioscopy is expertise-dependent.

And even among glaucoma specialists, it can be poorly reproducible.

We also know that gonioscopy for some patients can be uncomfortable, which leads to it being time-consuming, and this has generally discouraged non-glaucoma specialists from performing gonioscopy.

This issue was highlighted in another study that we conducted in 2024. Here, we used data from the Optum Healthcare Claims Database. This is an administrative claims database that provides us information on over 100 million Americans who are receiving care through the billing codes that are provided by their caring clinicians.

And what we found in this study is that in almost 200,000 patients who were undergoing initial glaucoma evaluation between 2009 to 2020, only 30% had a record of gonioscopy within six months of that first evaluation.

So this is in stark contrast to the 100% of patients who should be receiving gonioscopy according to the recommendations by the American Academy of Ophthalmology, but also the World Glaucoma Association.

We also found that the odds of gonioscopy were much lower in patients who are diagnosed with POAG compared to PACG, which suggests that some of these patients may have been misdiagnosed due to not having received gonioscopy, which highlights the importance of performing gonioscopy on all of our glaucoma patients since POAG and PACG have different practice patterns and treatment protocols.

Another key study in the field of angle closure was the Zhongshan Angle Closure Prevention Trial or ZAP Trial.

The primary purpose of the ZAP Trial was to study the effectiveness of laser peripheral iridotomy or LPI for preventing the progression of angle closure.

This study was a randomized controlled trial that was conducted in mainland China. And the primary objective of this study was to assess what is the risk of progressing from PACS to PAC over a six-year period.

The ZAP Trial yielded two primary findings, which are that LPI was effective. It reduced the risk of progression by around 50% over this timeframe.

However, the ZAP Trial also found that even when PACS was not treated, the risk of progression was very low at only 0.8% per eye-year.

And so this highlights a crucial limitation of gonioscopy, which is that the majority of patients who are identified with angle closure and being at risk for PACG actually do not go on to develop PACG, at least in a controlled study setting.

Next, I’d like to talk about anterior segment OCT or AS-OCT.

This is a relatively new technology that holds the promise of transforming how we evaluate patients with angle closure.

AS-OCT technology has been around for approximately 15 years at this point, and there’s been rapid evolution of the technology from earlier time-domain devices to more recent spectral-domain and swept-source devices.

The earlier devices tend to be slower, and the images tend to be of a lower resolution compared to the modern devices that are much faster and produce many, many more images in a shorter period of time.

This is a representative anterior segment OCT image. You can see that it beautifully portrays the anatomical structures in the front of the eye, including the cornea, the iris, and the lens.

However, one structure that you can’t directly visualize is the trabecular meshwork, and this is due to the wavelength of the AS-OCT devices, at least the swept-source AS-OCT devices.

Therefore, to detect angle closure, we have to rely on detection of the scleral spur, which is the anatomical structure that sits directly posterior to the trabecular meshwork.

Here is a representative image of an open angle. The yellow arrow shows the location of the scleral spur, and we can see here that the iris anterior to this point is not touching the inner corneal surface where the trabecular meshwork is located.

Therefore, in this image, the angle is open.

Here is a representative image of an angle sector with angle closure.

Again, this scleral spur is marked with the yellow arrow, and we can see that anterior to this point, there is contact between the anterior iris surface and the inner corneal curvature, which indicates the presence of angle closure or iridotrabecular contact, which defines angle closure in AS-OCT images.

In addition to assessing these images qualitatively, we can also perform a quantitative assessment to obtain measurements of different biometric parameters that describe the size, shape, and configuration of the anterior segment structures. In order to perform this type of analysis, we have to first segment the cornea, iris, and lens and mark the scleral spurs.

After doing so, we can then obtain these measurements of the biometric parameters that can be used for clinical care and scientific research.

This diagram shows some of the most commonly studied biometric parameters.

This includes the angle opening distance, which is the distance from the trabecular meshwork to the anterior iris surface. And this can be measured either 750 microns or 500 microns anterior to the scleral spur. Those distances were determined because the height of the trabecular meshwork is usually around 800 to 900 microns.

Other parameters that we tend to evaluate are anterior chamber depth, lens vault, which is how anterior the lens is compared to a line that is drawn from scleral spur to scleral spur.

Also important is anterior chamber width, which is the distance between the spurs, the pupillary diameter, which is the distance between the tips of the two sections through the iris here, as well as the anterior chamber area, which is the amount of area in the anterior chamber.

A few years ago, my lab asked this question: Are AS-OCT measurements predictive of intraocular pressure and angle closure outcomes?

And we have used various data sources to establish these relationships.

This was a study that we conducted in 2018 in which we looked at the correlation between intraocular pressure and angle configuration measured by OCT. Here, we used data from the Chinese American Eye Study, which was a population-based epidemiological study conducted here at the University of Southern California.

The AS-OCT device that we used was the Tomey CASIA SS-1000, which was a relatively new device at the time that could provide us with AS-OCT images.

This figure shows the correlation between intraocular pressure and measurements of trabecular-iris space area or TISA, which is the amount of area in the angle recess.

What you can see here on the right is that for most angle width measurements, there’s no correlation or relationship between intraocular pressure and angle width measurements.

However, below a specific TISA cutoff, there emerges a linear relationship, an inverse linear relationship between IOP and TISA measurements.

And this relationship suggests that as the angle narrows, when it reaches a critical threshold, then intraocular pressure is likely to rise.

More recently, I collaborated with Dr. Mingguang He and Dr. David Friedman, who were the lead investigators for the ZAP Trial, in order to study whether ocular biometric measurements could act as risk factors for predicting progression of primary angle closure disease.

And here in this study, we looked at 643 untreated PACS cases from the ZAP Trial so that we could observe the natural history of angle closure.

We required that participants have both gonioscopy and AS-OCT at baseline, and our outcome measure in this study was the progression from PACS to PAC.

This table summarizes our primary findings in the study. This is a multivariable logistic regression model, in which we assessed the association between angle opening distance or AOD with progression from PACS to PAC. And what we found was that after adjusting for age, iris curvature, and anterior chamber depth, a narrower AOD conferred a higher risk of angle closure progression. Specifically, for every 10 microns smaller in AOD 500, the odds of progression increased by 10%.

We also performed analysis in which we replaced AOD 500 as the measurement of angle width with gonioscopy score, which is a sum of the gonioscopy grades across all four quadrants.

And what we found here was that gonioscopy score was not associated with progression.

What this finding tells us is that AS-OCT provides us with unique information to forecast angle closure progression that is not possible using gonioscopy.

More recently, we’ve conducted additional follow-up studies to further establish the role of anterior segment OCT imaging for forecasting outcomes in angle closure eyes.

We conducted a similar study here in 2023, in which we studied LPI-treated eyes and whether biometric measurements could forecast progression. And we reached very similar conclusions that AS-OCT was much more predictive of angle closure progression in LPI-treated eyes than gonioscopy.

We’ve also performed studies where we looked at the role of AS-OCT measurements in the light and in the dark for predicting angle closure progression. And we found that actually the measurements in the light were more predictive, perhaps because most PACS cases are very narrow in the dark, and it’s actually the eyes that remain narrow in the light that are at higher risk of progression.

And the most recent study that we conducted, we looked at 18-month changes in biometric measurements and found that these short-term changes can forecast long-term progression, which highlights the importance of continuing to follow and image our patients over time.

And we’ve also used the ZAP Trial data to demonstrate that AS-OCT measurements can predict the amount of angle widening after LPI treatment, which could be useful for determining which patients should receive LPI versus monitoring or lens extraction.

So at this point, you’re probably wondering, how do I actually apply these findings when I’m caring for my patients? And here, I have two cases that I’d like to share with you that highlight the clinical applicability of these findings. So the first case is a 62-year-old Asian female who was seen in my clinic. She asked me, do I need laser for my eyes?

She works as a dentist, and she was seeking a second opinion about LPI. She was completely asymptomatic and not terribly motivated to have laser.

She had no relevant history.

And when I first saw her in September of 2021, she had best-corrected visual acuity of 20/20 in both eyes, pressures of 18 and 19. She was a low myope, and she had CCTs around 590.

On the exam, she was narrowed by Van Herick, and she had trace nuclear sclerotic cataracts in both eyes.

When I performed gonioscopy, I could not see pigmented trabecular meshwork in three of the four quadrants. But on indentation, all quadrants opened to scleral spur, and there was no evidence of PAS. On the undilated exam, she was a 0.5 cup in both eyes.

Here are the patient’s visual fields from both eyes, which we can see are completely full and intact.

And here are the OCT RNFLs.

We can see that there is borderline thinning superiorly, which likely reflects a temporal shift of the superotemporal RNFL bundles.

The rest of the RNFL is thick and intact.

The question that arises here is: does this patient need treatment?

Here, in order to better refine our care, we obtained anterior segment OCT images of both eyes.

You can see the AS-OCT scan of the right eye, which can be seen on the left. I’ve marked the scleral spurs with the orange circles. And we can see that anterior to this circle, there is no contact between the inner corneal surface where the trabecular meshwork is and the iris. Therefore, there’s an absence of iridotrabecular contact in both the right eye as well as in the left eye.

And here’s a zoomed-in view that shows that even though we could not visualize the pigmented trabecular meshwork, there was no iridotrabecular contact.

We also obtained an AOD 500 measurement of 0.13, which by itself is difficult to interpret.

Therefore, we turned to data from the ZAP Trial, in which we assessed the mean AOD 500 and determined that it was 0.09 millimeters.

And so you can see that this patient’s AOD 500 is greater than average, which is reassuring since we know that the narrower the angle on AS-OCT, the higher the risk of progression.

So based on this data, I diagnosed this patient as low-risk PACS, and we had a discussion regarding LPI. But since she wasn’t motivated initially and I didn’t think that she was at high risk, we elected to monitor annually. I did, however, give her acute angle closure precautions.

So we continue to monitor this patient, and I most recently saw her in May of 2025, and she was stable at that point.

The second case is of a 41-year-old Caucasian gentleman who was coming to me asking, do I have narrow angles? He works as an investment banker, and he was referred by his optometrist for an angle evaluation.

He was also asymptomatic.

He had a positive family history of narrow angles, and his mother had received LPI, but she was actually rather unhappy due to developing dysphotopsias afterward.

At baseline evaluation in 2018, this patient had a best-corrected visual acuity of 20/20 in both eyes. His pressure was 17.

He was a very low hyperope, and his CCT was around 580.

On slit lamp exam, he was also narrowed by Van Herick, and his lens was completely clear.

On gonioscopy, I could not see angle structures in three of the four quadrants, and I could see non-pigmented TM in the inferior quadrant. And on indentation, I could still only see pigmented TM in the inferior quadrant.

I did not see any PAS, however.

On undilated fundus exam, he had a cup-to-disc ratio of 0.4 in the right eye and 0.5 in the left eye.

Here are his visual fields, which are also completely full and intact in both eyes.

And here are his OCT RNFLs.

We can see that both have great signal strength and that his OCT RNFLs are thick and healthy in both eyes, although there is some relative superior thinning in the left eye. His GCC was also full.

So at that time, I asked myself, does this patient need treatment?

And, again, we turned to anterior segment OCT to better refine our management.

Here, the scleral spurs are again marked using the yellow dots, and we can see that in the right eye in the horizontal section, which is the temporal-nasal section, there is iridotrabecular contact, and that this is also true in the inferior quadrant as well. In the left eye, similarly, there is extensive iridotrabecular contact superiorly and in the horizontal section both temporally and nasally.

So this degree of angle narrowing appears more severe on imaging than our first patient.

To contextualize the AOD 500 measurement, which was 0.04 in both eyes for this patient, we find that this patient is quite a bit below the average of the ZAP Trial, which was 0.09.

And so, given the AS-OCT findings, I diagnosed this patient as a high-risk PACS case, and I recommended LPI in both eyes. However, this patient deferred LPI due to concern about dysphotopsias, which he had heard about from his mother. And so I gave him acute angle closure precautions and decided to follow him every nine months.

Several years elapsed, and he was originally stable. But then in 2023, he presented now with elevated intraocular pressure to 24 in both eyes.

The remainder of his exam was largely unchanged.

At this point, we repeated the OCT RNFL. Fortunately, there was no difference in RNFL thickness, which suggests that we caught the elevated IOP early enough before any glaucomatous damage occurred.

Here are his AS-OCT images from that day in 2023, and we can see that his AOD 500 was very, very narrow at that point, at 0.03. We can see that he has very prominent iridotrabecular contact in both the nasal and temporal sectors.

And here is his other eye in which you can see that there is also extensive iridotrabecular contact, with an AOD 500 of 0.04.

So given that he now has elevated intraocular pressure, he fits the definition of primary angle closure.

And the consensus here by both the World Glaucoma Association and the American Academy of Ophthalmology is that some form of angle-widening treatment is indicated at this point.

There are two primary forms of treatment for angle narrowing, the first being LPI and the second being lens extraction. And there is excellent evidence from both the ZAP Trial as well as the EAGLE Trial that both forms of treatment are very effective at widening the angle and lowering intraocular pressure.

So specifically from the EAGLE Trial, we know that PACS with high pressures over 30 millimeters of mercury are highly responsive to LPI treatment. Here, even though his IOP was not that high, LPI was a viable treatment option. And so we went ahead and performed LPI at this point. He was more amenable since he had high eye pressure. And, after the LPI, we can see that in the right eye, his AOD 500 went from 0.03 to 0.12 millimeters and that there was resolution of the iridotrabecular contact, particularly here in the temporal sector.

He also responded to treatment in the left eye, although the response was not as dramatic, where he went from an AOD 500 of 0.04 to 0.08. And, again, we can see that the lengthy ITC that he had previously was reduced to just a little bit at the iris root.

So in terms of his clinical course, after LPI treatments, his pressure responded and went back to his baseline, 18 in the right, 17 in the left. And I counseled him about the possible need for clear lens extraction given that, especially in the left eye, he still was somewhat narrow. And we continue to follow this patient every six months, closely, in case his pressure goes up again.

At this point, I’d like to transition to talking about artificial intelligence or AI and its role in anterior segment OCT imaging and angle closure care.

AI is obviously a transformative technology that promises to change many aspects of our daily lives in the society that we live in.

AI has become a powerful tool in our day-to-day activities, through large language models such as ChatGPT. We also know that the introduction of AI can be disruptive to the global economy, due to our reliance on the technology as well as on the GPUs that power this technology.

And time will tell whether AI is more friend or foe. Obviously, there is a lot of concern and apprehension about the capabilities of AI and the potential for its misuse in the future.

However, here we’re talking about AI and healthcare, and the implications of the technology are quite clear. We hope that AI can help us improve access to care, provide more reproducible and equitable care, and hopefully lower the cost of care, especially as the number of patients that we have to take care of is rapidly increasing. And so in order to harness this technology, which is just a tool, we have to identify clinical gaps and apply this technology to fill those gaps.

So one gap that we used AI to fill is the fact that most clinicians don’t have the time and most also don’t have the expertise to manually analyze AS-OCT images in their busy clinics for the care of their angle closure patients.

An important step in the analysis of these images is identification of the scleral spur, which is not a common part of ophthalmological training at this point.

So one question that we asked a few years ago is: Can AI be used to automate the analysis of AS-OCT images to detect gonioscopic angle closure?

And here, we used data from the Chinese American Eye Study, again, to develop an AI model for automated detection of gonioscopic angle closure. The way that we developed this model is that we took about 3,400 AS-OCT images that were paired with gonioscopy grades provided by a human clinician who had performed manual gonioscopy on all of these patients as part of the Chinese American Eye Study. And using a convolutional neural network, which is a form of artificial intelligence, we trained this model to essentially translate an AS-OCT image into a gonioscopy grade. And then grades zero and one were classified as angle closure, due to inability to visualize the pigmented trabecular meshwork, and grades two, three, and four were deemed to be open.

And this is the figure that shows the performance of this model or algorithm.

This is called a receiver operating characteristic curve or ROC curve, and it plots the sensitivity of the model versus one minus specificity.

And so the perfect model would have an area under this curve or AUC value of one, which would look like a box, but this type of performance is not possible in a real-world environment. And so this model, for detecting angle closure in a single quadrant using a single AS-OCT image, achieved an area under the curve of 0.93.

That’s this red curve, which corresponds to a balanced sensitivity and specificity of around 85%, which is quite good.

Given that we typically detect angle closure in multiple quadrants to define primary angle closure spectrum, we also looked at the performance for detecting angle closure in two or three quadrants, and the performance here was even better with an area under the curve exceeding 0.95.

So one issue with AI to keep in mind is that oftentimes, the performance of these algorithms is quite fragile when you take them and try to apply them in settings outside of the original training setting or population.

And so to assess whether this algorithm could be used in other clinical settings, we performed a generalizability study in 2021.

We were loaned data from the polyclinics run by SingHealth in Singapore, and we also acquired our own data at the USC Roski Eye Institute.

And so you can see here now there are three ROC curves, and the area under the curve for each of these different populations was very similar, at around 0.89 to 0.92. So this is despite differences in patient populations and clinical settings.

It is important to note, though, that we used the same AS-OCT system across all three sites.

It’s also important to note here that in our cohort at USC, these patients actually received gonioscopy by two clinicians.

And so you can see the agreement here between the two clinicians is marked by the sensitivity and specificity of the second clinician compared to the first clinician. And we can see that the AI model achieves the same level of performance, which suggests that it is performing at a very high level.

So the next question that we asked is: Can AI be used to automate scleral spur detection and biometric analysis, given that this appears to be the future of angle closure care?

And in 2020, we developed an AI model for our original CHES dataset, in which we could identify the scleral spur at an expert level of performance. However, here, we ran into a generalizability issue, which is that trying to apply this for an FDA-approved OCT system such as the Heidelberg Anterion, we found that this algorithm no longer performed as well. And so in 2023, we conducted a study in which we developed a new version of this algorithm specifically for the Anterion using data that was collected on this device.

And in this study, we wanted to demonstrate that we were achieving expert-level scleral spur detection and that the measurements corresponding to these scleral spur locations were reliable.

This diagram shows how we developed the model. Here, we used just over 4,000 images, and the images were paired with the scleral spur coordinates.

And the model relied on a convolutional neural network architecture.

It’s important to note here that the training data was acquired in Germany, and there, they have a very different patient population from what we have here in Los Angeles and in the United States. And so we wanted to perform an independent validation of this algorithm.

And the way we did that is we collected 1,300 images that yielded 2,600 scleral spur locations from patients at the USC Roski Eye Institute. The scatterplot shows the human-human agreement, which is the agreement between a reference grader and a glaucoma expert. And we can see that the error by the glaucoma expert, calculated as the reference minus the predicted scleral spur coordinates, was around 60 microns.

This scatterplot shows the agreement between the human and the AI algorithm.

And we can see here that the scatterplot looks very similar and that the median error is around 56 microns. So at least in terms of scleral spur detection, the AI model simulates a glaucoma expert very closely.

This table shows the inter-rater reproducibility of the measurements that are associated with those scleral spur locations.

In the column on the left, you can see the agreement between the reference rater and the glaucoma specialist. And here, we’ve calculated intraclass correlation coefficients.

These ICC values can range from zero to one, with one being perfect correlation.

And we can see here that all of these ICC values exceed 0.96, which is very high correlation between the two humans.

However, this level of correlation is matched by the AI or deep learning model, which achieves similarly high ICC values. So the takeaway here is that with a simple click of a button, these AI algorithms enable us to access expert-level measurements that can be then applied for the clinical care of our angle closure patients.

So let me end by just talking briefly about the future of AS-OCT imaging and angle closure care.

Ideally, what we would like is to complement our gonioscopic evaluation with anterior segment OCT so that we have a more precise risk stratification of our PACS patients. Then we can deliver early treatments to those high-risk patients and ultimately prevent PACG and PACG-related blindness.

Currently, in order to understand the long-term clinical implications of AS-OCT measurements, we’re conducting the Multiracial Angle Closure Progression Study or MAPS.

This is a study that is funded by the National Eye Institute, and we’re conducting it here in Los Angeles to take advantage of our very diverse patient population.

This study has three primary aims. The first aim is to elucidate racial differences in primary angle closure glaucoma prevalence. We know that PACG is most common in Asia, but it’s also very common amongst Hispanics and Black Americans.

The exact cause is not known. And so what we’re doing is that we are screening 4,000 participants, and we’re deriving normative data to help understand the differences in anatomical configurations between people of different racial and ethnic groups. And we will then analyze this to try to correlate this with what we know about disease prevalence.

We’re also trying to establish OCT-based definitions of what is a narrow angle? I get this asked all the time. And at this point, we don’t have a strict cutoff or OCT-based finding that we can use to define a narrow angle that merits treatment.

So here, we’re taking 300 high-risk cases. These are patients with the most narrow angles who were evaluated in aim one, and we’re going to be monitoring them over the course of three years.

And our endpoints here will be elevated intraocular pressure over 21 millimeters of mercury or the development of glaucoma.

And so we hope by using these clinically meaningful endpoints that we can begin to interpret what it means to be narrow on AS-OCT.

And finally, we also know that we don’t have great tools for identifying patients who are at high risk for acute angle closure, and even our ability to predict the progression from PACS to PAC is somewhat limited. And so in the third aim, we are leveraging a unique device that we developed that can dynamically modulate the activity of the iris, in order to understand the role of these dynamic changes as risk factors in angle closure.

So here’s an example of what this looks like.

The video shows an AS-OCT feed of a patient who’s sitting in the dark. And then you can see when the light comes on, the pupil constricts.

And when the light is turned off, then the pupil dilates again.

So what the machine does is it calculates the difference between the fully dilated state and the fully constricted state, and it calculates the mid-dilated pupil size.

The reason why we want to hold the iris in a mid-dilated state is because we know that patients with acute angle closure often come in with mid-dilated pupils. So is there something unique about this pupillary state that increases pupillary block and perhaps triggers acute angle closure? So now the light has come back on. It’s quite dim, and you can see that the pupil has now assumed a mid-dilated state.

So using tools such as this, we hope to be able to precisely characterize the activity of the iris and correlate this with the clinical outcomes of the patients over the three years.

So just to summarize this talk, I want to emphasize that gonioscopy really is essential for evaluating the angle. It is often the most convenient and cost-effective tool that we have for identifying patients with angle closure who might be at risk for PACG. And therefore, it is still very, very important that we all master this technique and continue to practice on our patients.

It’s also important to note that PACG is often undetected until blindness has occurred. And so it becomes increasingly important to keep that in mind as we are evaluating patients, using gonioscopy or AS-OCT.

Recent research supports that AS-OCT can help refine assessment of angle closure. And as more and more research is conducted, we hope that we are better able to deliver this type of precision care in angle closure eyes.

And, hopefully, these AI tools will become more widely available, and they help support our evaluations using AS-OCT and make this a more convenient tool in our clinical practices.

However, I do want to emphasize that widespread adoption of AS-OCT imaging for the care of our angle closure patients requires longitudinal studies such as MAPS before we can fully interpret what these measurements mean and how we can use them to guide treatment.

So I’d just like to thank all of you for your attention today. I hope you learned something about gonioscopy and AS-OCT imaging and the future of angle closure care. I’d like to acknowledge the hard work and contributions by the many members of my lab who contributed to the work that I spoke on today. And if you’re interested in learning more about our research, I invite you to follow us on X.

So, thank you again, and I hope that you have a wonderful day.