The visual system of humans is not fully developed at birth and gradually improves during the first few years of life. World Health Organization estimates there are about 19 million children in the world suffering from visual impairment. However, 70–80% of them could prevent or treat it, if they have been detected with visual disorders in childhood.
Using an eye tracker with the patient, we could know which point of the screen that the patient is looking at, but analyzing gaze data of the patient is quite complicated, that is why we choose machine learning to help ophthalmologists to diagnose visual pathologies. We could collect data by performing various tests on the patients, and the ophthalmologist supervises this operation.
Thanks to DIVE, a visual examination device, we have lots of gaze data from children. We used that gaze data to feed into the ML model, and some tests also made use of the ASA (Accelerated Stochastic Approximation) algorithm to provide meaningful constants before using in the ML model. Outputs of the model are the probability of abnormalities for each eye and each pathology.