Feasibility of EEG-based machine learning for the objective assessment of non-Strabismic binocular vision dysfunction

With the increasing prevalence of prolonged near work, non-strabismic binocular vision dysfunction (NSBVD) has become a growing concern. Current diagnostic methods primarily rely on subjective symptoms and time-consuming examinations, highlighting the need for objective physiological markers. This pilot study explores the feasibility of utilizing electroencephalography (EEG) combined with machine learning as an objective, auxiliary approach for NSBVD assessment. We analyzed EEG activity in 15 NSBVD patients and 15 healthy controls during a natural viewing vergence task. Time frequency and topographic analyses were used to identify neural features associated with vergence insufficiency. The groups exhibited distinct neural patterns. Healthy controls showed strong activation in visual areas, whereas NSBVD patients displayed reduced activity, coupled with compensatory increases in frontal activity, particularly in the theta and alpha bands. A linear support vector machine (SVM) trained on these features achieved 76.67% accuracy (AUC = 0.87). These findings suggest that specific neural patterns may serve as potential biomarkers for binocular dysfunction. This study demonstrated the feasibility of objective screening, though validation in larger cohorts is needed for clinical use.