[Reading Science] AI Detects Heart Disease and Parkinson's Disease Just by Examining the Retina

UK Research Team Develops Self-Directed Learning AI
"Generative AI Method Shows Excellent Performance with Few-Class Data"

An artificial intelligence (AI) has been developed that can observe changes in the retina of the eye to examine not only ophthalmic conditions but also serious diseases such as heart disease and Parkinson's disease more affordably and accurately.

Retina. Photo by Pixabay

On the 13th (local time), a research team from University College London in the UK published a paper containing these research results in the international academic journal Nature. There have been tools that interpret diseases using retinal images before. However, the AI called 'RETFound' developed by the research team is special in that it introduced a method called self-supervised learning. In other words, there is no need to analyze each of the 1.6 million retinal images used to train the AI to classify them as normal or abnormal. This process requires enormous time and cost. The research team developed an AI that can significantly save time and cost by using a method similar to training large language models (LLMs) like ChatGPT.

ChatGPT and LLMs predict the next word in the context of previous words by utilizing numerous existing texts created by humans. The AI developed by the research team works in a similar way. It uses many existing retinal images to predict what physical condition changes in newly added retinal images reflect.

The AI developed by the research team showed higher performance than other AIs even with learning from a small amount of labeled data. For example, it was trained using retinal image data from 100 Parkinson's disease patients and 100 normal individuals, but through its unique generative AI-style prediction, the accuracy of interpretation was higher. In the case of diabetic retinopathy, the prediction accuracy ranged from 0.822 to 0.943 (closer to 1 means more accurate), which was much higher than existing AI models. Predictions for heart attack, heart failure, stroke, and Parkinson's disease were also relatively accurate.

The human retina is regarded as a 'window' to look into health. This is because it is the only part of the body where the capillary network composed of very small blood vessels can be directly observed. Pierce Keane, an ophthalmologist at Moorfields Eye Hospital in London, UK, explained, "If there is a systemic cardiovascular disease such as hypertension that potentially affects all blood vessels in the body, it can be directly confirmed through retinal images." The retina is also the terminal of the central nervous system connected to the brain. Observing the condition of the retina well means that the health of brain nerve tissue can be assessed.