by Connor Courtien, RDPFS Intern
Recent developments in applying artificial intelligence (AI) to ophthalmological medical care, utilizing a process called deep learning (DL), show promise when it comes to diagnosing several eye diseases. Diagnosing eye diseases through a DL model largely began with diabetic retinopathy, with the first such AI-enabled ophthalmic device approved by the Food and Drug Administration (FDA) in 2018. Over the past five years, more and more eye diseases are being diagnosed in this way, including macular degeneration and glaucoma. Some examples of effective AI diagnosis covered in the American Journal of Managed Care (AJMC) include two models for diagnosing keratitis and keratoconus, conditions affecting the cornea, which were 99.9 percent and 99.6 percent accurate, respectively. Models such as these are already very precise and will only become more so as the technology develops, allowing for clinicians to utilize AI-enabled tools for more efficient and accurate diagnosis than they could provide manually. DL models can diagnose diseases through a training process where they learn to infer the underlying pattern shared by each image of a common category, such as healthy retinas versus those with diabetic retinopathy. This allows for predicting the category where a new image belongs and is accomplished by providing a large number of images that are labeled as belonging to one category or another. From these images, over several iterations known as “epochs,” the model can learn to distinguish one category from another. Consequently, the number of images, as well as their quality, greatly increase accuracy, and these are often the constraints when trying to build a model for diagnosis. To learn more about the current state of AI-enabled eye disease diagnosis, check out this recent AJMC article on how Improvements in AI Make Ocular Surface Disease Diagnosis Possible and this brief academic publication from the National Library of Medicine on Artificial Intelligence in Eye Disease: Recent Developments, Applications, and Surveys.
