Artificial intelligence (AI) is beginning to revolutionize the everyday practice of ophthalmology. When an AI system gains the ability to recognize patterns or markers of a disease, it can then become a tool for automated diagnosis. AI systems are already available or in development for the detection of multiple ophthalmic diseases, including diabetic retinopathy, age-related macular degeneration and glaucoma.
At SCEI, our researchers are exploring the use of large datasets and artificial intelligence to improve the diagnosis and treatment of vision disorders. Our priority is refining and advancing this important tool to bring more advanced treatment options to more patients.
SCEI Director Rohit Varma, MD, MPH, recently outlined four key points about the value of artificial intelligence in Ophthalmology Times:
1) AI is gaining diagnostic credibility. It has the potential to offer a higher level of objectivity and precision in analyzing medical images and formulating prognoses than many human physicians. Supporting this potential is the widespread use of Big Data and increased computing capacity.
2) AI is becoming a trusted medical partner. The new AI can offer a cogent explanation of why it makes a certain diagnosis, and can point to the precise part of a scan that led to its conclusions. This capability is a major step forward.
3) AI saves time. Reports show that AI can perform repetitive analytical work in seconds — the type of work it might take humans hours to do. Also, AI can assemble 3-D models of an ophthalmologic tumor from data taken from hundreds of 2-D scans.
4) AI can increase access to care. Patients living in remote areas or away from specialists will be able to use the AI version of telemedicine to have eye exams and consultations. The data collected can then be sent to an AI center for analysis.
Deep Learning and Diabetic Retinopathy
In 2018, Varma led a study of a deep learning system (DLS), a machine learning technology with the potential for screening eye diseases.
The Goal: The study was set-up to determine how the DLS would perform in identifying referable and vision-threatening diabetic retinopathy (DR), glaucoma and age-related macular degeneration in multiethnic populations with diabetes.
The Challenge: In the past, a few deep learning systems showed promise, but the studies were done using databases from homogenous populations of white individuals. To truly understand the DLS potential, it had to be evaluated using retinal images of patients of difference races and ethnicities — and varying fundi pigmentation (the fundus is the interior surface of the eye).
The Process: A DLS was trained and validated to detect the eye diseases using retinal images. Varma and his research team employed images from an ongoing community-based diabetic retinopathy (DR) screening program in Singapore, as well as 10 additional multiethnic datasets from different countries with diverse populations with diabetes.
The Findings: The DLS had a sensitivity of 90.5 percent and specificity of 91.6 percent for detecting referable DR;100 percent sensitivity and 91.1 percent specificity for vision-threatening DR;96.4 percent sensitivity and 87.2 percent specificity for possible glaucoma, and 93.2 percent sensitivity and 88.7 percent specificity for age-related macular degeneration, compared with professional graders.
The Conclusion: DLS has high sensitivity and specificity for identifying diabetic retinopathy and related eye diseases.
Further research is needed to evaluate the applicability of the DLS in health care settings and its utility in improving vision outcomes.
Read more about deep learning and screening for diabetic retinopathy.