AI revolutionizing AMD diagnosis, treatment
As COVID-19 ushered telemedicine and virtual visits to the forefront of clinical practice, artificial intelligence has emerged as an important way to augment care.
In this arena, ophthalmologists appear to be ahead of the curve. For the past few years, researchers have been exploring improvements in imaging technology as well as delving into the use of deep learning and machine learning algorithms to enhance patient care.
In an interview with Healio, Rohit Varma, MD, MPH, CMO at CHA Hollywood Presbyterian Medical Center and CEO at Southern California Eye Institute, discussed the current role of AI in ophthalmology for conditions such as age-related macular degeneration, the effect of COVID-19 on development in this space and what hurdles remain.
Healio: How is AI currently being used in ophthalmology?
Varma: Currently, AI is used mainly for detection of disease across various areas of ophthalmology, including keratoconus, diabetic eye disease, glaucoma and AMD. However, it is beginning to be used for detecting and predicting systemic diseases such as atherosclerosis and stroke as well as identifying which people have high blood pressure and which people may go on to develop diabetes. We are on the cusp of using AI for prediction as well as detection in regular practice, but for now, we are mostly using AI for automated detection of various eye diseases.
Healio: In what ways are clinicians seeking to use AI for diagnosis or management of AMD? How would these improve upon current practice?
Varma: As I mentioned, AI is mainly being used for detection of AMD — drusen, choroidal neovascular membranes, hemorrhages and so on — as an adjunct to the clinician’s evaluation and judgment. AI has not replaced the clinician, but it has become an additional tool that can validate what the clinician sees on fundus or OCT images of the retina. Also, at times, the clinician may miss some something on examination of the OCT or the fundus or because they have not examined all the fundus or OCT images carefully. In these situations, AI can identify what was missed and bring it to the attention of the clinician, so we are less likely to make diagnostic errors.
Healio: What studies have been conducted on the use of AI in patients with AMD?
Varma: There is a whole series of studies that have been conducted on the use of AI in AMD. The main goal of these studies was accurate detection of AMD in the eye. Beyond that, however, studies have also been done to evaluate relationships between structure and function in the eye. Specific to AMD, this means looking at structure of the retina and its relationship to visual fields and electroretinogram findings as well as trying to assess, correlate and even predict what the visual acuity will be. These studies have also been looking at structure of the retina and potential prediction of patient-reported outcome measures, using vision-related quality of life instruments. Essentially, studies are no longer focusing on use of AI for a one-time baseline assessment for detection. Now, the question is can AI help predict what is going to happen to the vision or function of the eye or the patient? Can AI predict what is going to happen with changes in the retina? Can AI predict what is going to happen to changes in the visual field? The current studies are focusing on these questions.
Healio: What effect has the COVID-19 pandemic had on development in this space?
Varma: It has had a huge effect in that it has accelerated the pace of development. We have seen advances in imaging — in what is needed to image the eye — as well as advances in using large datasets for analyzing images and clinical data to enhance our ability to predict the course of disease. Under normal circumstances, this would have taken many, many years. Instead, clinicians and scientists are realizing that, despite the pandemic, we still need to provide care for our patients. COVID-19 accelerated both the development of analytic techniques and the clinical research so that we could bring the predictive capability of AI to the patient and provide better care without necessarily needing to see these patients in person. This is so important because, as is often true in ophthalmology, early detection and management of the disease is critical because if they are managed late, many of these conditions cause irreversible vision loss. For instance, AMD is one of the leading causes of blindness in both the U.S. and around the world. Earlier detection, and subsequently earlier initiation of treatment, can prevent people from going blind.
Healio: What improvements in technology have allowed you and your colleagues to take advantage of AI?
Varma: There have been significant advances across the board, from imaging to analysis to the development of new algorithms. First, there has been improvement in our ability to obtain more detailed images of the eye as well as the ability to obtain better images of the eye using remote systems. Second, we have developed better machine learning algorithms and deep learning algorithms to detect and predict disease. Third, we have larger amounts of data and better statistical tools that have allowed us to test and validate these algorithms using various subsets within these large datasets.
Healio: What challenges might clinicians face with the use of AI?
Varma: First, we’re still trying to develop better tools so that patients who are at home or in remote areas or cannot come into the office, the physician can obtain deeper and clearer images of the eye. Second, we still need better data on how well we can utilize AI for prediction. Lastly, and most importantly, we need to get this information out to various clinicians so that they become aware that these tools exist and then implement them into their clinical practice. Many physicians are fairly conservative people by nature and usually not “early adopters” of novel technology and treatments. Physicians are trained in specific ways, so they know that what they already learned works. Introducing new technology and new ways of detecting, predicting and treating people is always a challenge. Perhaps the younger generation of physicians who are more tech savvy and willing to adopt technology in their treatment algorithms will help accelerate this process.
Healio: Are there new technologies that are being studied?
Varma: There is a lot of new technology with respect to imaging. Specific examples include looking at OCT angiograms and detecting changes in blood flow or looking at the function of specific layers of the retina and how they are performing and how their function is changing as disease develops. Those tools are absolutely critical and are continuously evolving, which is very exciting.
In the last 10 to 15 years, OCT has pretty much revolutionized the treatment of eye conditions, from glaucoma to diabetic macular edema to AMD, and we are getting to a point where we’re moving toward deeper assessment of disease mechanisms that we have so far only been able to assess purely from a structural standpoint. We’re now going to be able to assess those same layers of the eye to see not just whether their structure has changed but also whether the cells in those layers are functioning better or whether they are diseased. Once we can do that, we will be able to predict the course of disease very early on and potentially treat and prevent it.
The most exciting aspect of this for me will be when we reach a point where people can use these tools remotely and we can utilize AI to begin to study millions more people and subsequently reduce the overall burden of people going blind from AMD and other eye diseases.
Healio: Do you have anything that you would like to add?
Varma: I would like to add that with the use of AI, we have been able to look at the eye and predict other diseases, such as heart attacks, strokes, high blood pressure and diabetes. We’re even studying whether we can predict diseases like Alzheimer’s. So, the eye’s ability to tell us about the health of the entire body is another exciting area to watch, and I hope we can continue to build on that as well.