Eye movement behavior identification for Alzheimer's disease diagnosis

We develop a deep-learning approach to differentiate between the eye movement behavior of people with neurodegenerative diseases during reading compared to healthy control subjects. The subjects with and without Alzheimer's disease read welldefined and previously validated sentences including high- and low-predictable sentences, and proverbs. From these eyetracking data trial-wise information is derived consisting of descriptors that capture the reading behavior of the subjects. With this information a set of denoising sparse-autoencoders are trained and a deep neural network is built using the trained autoencoders and a softmax classifier that identifies subjects with Alzheimer's disease with 89.78% accuracy. The results are very encouraging and show that such models promise to be helpful for understanding the dynamics of eye movement behavior and its relation with underlying neuropsychological processes.