Predicting Adolescent Visual Acuity Using Transformer-Based Time Series Analysis

Visual acuity prediction in adolescents presents a significant challenge in ophthalmology, particularly for myopia management. This paper introduces Visual Acuity Sequential Transformer (VAST), a novel Transformer architecture specifically designed for predicting visual acuity progression using longitudinal clinical data and generative temporal encoding. VAST achieves Pearson correlation coefficients of 0.970 (OD) and 0.969 (OS) for spherical equivalent, and 0.981 (OD) and 0.979 (OS) for axial length measurements. Extensive analysis demonstrates that the VAST model can effectively capture complex progression patterns while adapting to different developmental stages and varying amounts of historical data.Clinical relevance- VAST provides clinicians with a powerful tool for early detection of rapid myopia progression and personalized treatment planning. By accurately predicting visual acuity changes, particularly in critical developmental periods, this model enables more timely interventions and better-informed clinical decisions, potentially improving long-term vision outcomes in adolescent patients.