Predicting Adolescent Visual Acuity Using Transformer-Based Time Series Analysis.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Visual acuity prediction in adolescents is challenging, especially for myopia management.
• Accurate prediction of myopia progression is crucial for timely intervention.

Purpose of the Study:

• To introduce Visual Acuity Sequential Transformer (VAST), a novel AI architecture for predicting visual acuity progression.
• To evaluate VAST's performance in capturing complex myopia progression patterns using longitudinal clinical data.

Main Methods:

• Developed VAST, a Transformer architecture utilizing generative temporal encoding.
• Trained and validated VAST on longitudinal clinical data from adolescent patients.
• Assessed model performance using Pearson correlation coefficients for spherical equivalent and axial length.

Main Results:

• VAST achieved high accuracy with Pearson correlation coefficients of 0.970 (OD) and 0.969 (OS) for spherical equivalent.
• The model demonstrated excellent performance for axial length measurements, with coefficients of 0.981 (OD) and 0.979 (OS).
• VAST effectively captured complex progression patterns across different developmental stages and data availability.

Conclusions:

• VAST is a powerful tool for clinicians in managing adolescent myopia.
• The model enables early detection of rapid myopia progression and supports personalized treatment strategies.
• Accurate visual acuity prediction by VAST facilitates timely interventions, potentially improving long-term vision outcomes.