Using Artificial Intelligence to Detect a Disease
Sayan Saha
David Thompson Secondary
Floor Location : M 111 N

Diabetic retinopathy is an eye disease which affects many patients with long-term diabetes. If the disease is not detected in time, it may lead to vision impairment and eventually blindness. Progression to vision impairment due to the disease can be slowed or averted if it is detected in time, but it is often difficult to do so as patients rarely show any symptoms in the early stages of diabetic retinopathy, and when symptoms do begin to appear, it can be too late to provide proper treatment. Patients are generally tested for the disease only when symptoms begin to show since detecting the disease is a very manual process which involves a trained clinician carefully examining photographs of a patient’s retinas to look for signs of the disease.
The purpose of this project was to design an automated system for detecting the disease using photographs of a patient’s retinas so that the eyes of diabetes patients who are at risk of getting the disease could regularly be tested for diabetic retinopathy, even before symptoms began to show, and if the disease was detected, proper treatment would be provided. The automated system was created using a neural network, a form of artificial intelligence which can learn to classify images. To train the the neural network to detect the disease, thousands of retinal images, some with diabetic retinopathy and some without, were shown to it so that the network could learn to differentiate between a normal retina and the retina of a patient with diabetic retinopathy. The neural network was then tested on over 53,000 retina images which it had never seen before, some with diabetic retinopathy and some without, and the neural network was able to predict whether or not the retinas in the images had diabetic retinopathy with nearly 90% accuracy.
This neural network can now be easily used by optometrists to regularly monitor the eyes of patients who are at risk of acquiring diabetic retinopathy.