Scientists Develop AI-Based Systems to Diagnose Rapidly Advancing AMD
Scientists supported by National Eye Institute (NEI) are developing artificial intelligence/machine learning (AI/ML)-based systems that perform two functions. Firstly, these systems screen for AMD (age-related macular degeneration) and secondly, predict which patients are likely to progress to late AMD within two years.
The systems also evaluate the risk of an individual separately for developing late wet AMD (neovascular) from the risk of developing late dry AMD (geographic atrophy).
“This is important since treatment approaches for the two subtypes of late AMD are very different: Neovascular AMD needs prompt diagnosis, since delay in access to anti-VEGF therapy is usually associated with a poor outcome,” Chew said.
Such systems are autonomous and provide clinical and diagnostic information without expert interpretation. They are created by training a deep convolutional neural network on large datasets that contain thousands of patients with known outcomes. The neural network extracts features from the retinal images of these patients, including the size and type of drusen and pigment abnormalities. The imaging data is combined with patient information like age, smoking status, genetic profile, and eye disease outcome. By analyzing this data, the network can identify patterns that assist with screening and prediction.
iHealthScreen
iHealthScreen is an AI/ML-based system developer. Their iPredict system helps non-eye care specialists to screen patients for AMD and identify individuals with early AMD who are at risk for accelerated disease and vision loss.
The iPredict system uses a fully automated fundus camera to take color images of a patient's retinas, along with other parameters, and transmits them to a centralized server for analysis by an AI/ML-based algorithm. The system generates a report within a minute, classifying the patient as referable or non-referable for AMD, and providing recommendations to visit an ophthalmologist if necessary.
For those found to be referable, the system generates a prediction score of the individual's risk for developing late AMD within one or two years. The iPredict system has been prospectively validated and submitted to the FDA for clearance.
NIH Developed AI-Based Late AMD Prediction
The NIH system and iHealthScreen were both developed using images from AREDS participants, with the former being trained and validated on an expanded set of images from participants with intermediate AMD in both eyes from the AREDS2 trial.
Both systems detect reticular pseudodrusen, a type of lesion that causes a spotted pattern in the macula, but they differ from typical drusen and can be difficult to detect. Though there is some evidence that reticular pseudodrusen may indicate a higher risk for late disease progression, further research is needed to understand their underlying biology and prognosis.
