Application of Unsupervised Learning for Eye-Tracking Signals Analysis

This study investigates the diagnostic potential of oculographic signals using unsupervised machine learning methods, specifically focusing on comparing the diagnostic capacity of different features extracted from eye movement data. The aim of the study was to determine which features are more effective in discriminating between groups of patients with post-stroke visual impairment, especially those with diplopia, nystagmus, and neglect. The study included 10 healthy subjects and 101 patients with stroke and other neurological impairments undergoing rehabilitation. Oculographic data was collected under static and dynamic conditions using a C-Eye Pro device with an eye tracker. The analyzed features included target hit statistics, angular deviation, and features obtained using two algorithms: angular velocity and interval variable threshold. This study contributes to the understanding of oculographic signal analysis and provides insights into which features may be most effective in detecting specific eye movement disorders. The results may help develop more accurate diagnostic tools in clinical settings for earlier detection and rehabilitation.