Digitizing Binocular Vision Diagnostics via EyeQ Platform: A Study on Algorithmic System Architecture for Convergence Insufficiency (CI) Detection

Abstract

Background: Convergence Insufficiency (CI) is a complex neuromuscular visual disorder traditionally diagnosed using manual, subjective instruments like the RAF Rule, which heavily rely on examiner estimation and patient response times. Methods: To address these clinical limitations, this study introduces the EyeQ platform, an innovative diagnostic framework that integrates eye-tracking technology with deep learning architectures powered by the TensorFlow framework. A single-blind comparative study was conducted involving 50 participants. A broad, heterogenous age spectrum (ranging from 6 to 82 years) was intentionally selected to rigorously validate the system's algorithmic adaptability and robustness across diverse demographic profiles and age-related physiological ocular variations. The platform utilizes a dynamic "Approach-Recede" mechanism to monitor ocular neuromuscular responses and isolate precise Break and Recovery Points. Automated diagnostic outcomes were benchmarked directly against independent clinical evaluations performed by a certified optometrist.  Results: The EyeQ platform demonstrated high diagnostic efficacy, achieving an overall accuracy of 94% by matching the specialist's clinical findings in 47 out of the 50 cases. Crucially, detailed clinical analysis revealed that the 6% statistical variance (3 cases) was entirely attributed to physical anatomical obstructions—specifically, two cases of ptosis and one case of severe eyelid edema—which occluded the digital region of interest (ROI) and hindered feature extraction,

rather than systemic or algorithmic failure. Conclusion: The EyeQ system provides exceptional digital objectivity and reliability, eliminating subjective examiner bias and establishing a standardized digital database for longitudinal vision tracking. Given its high diagnostic precision and architectural consistency, the platform is highly qualified to It shows potential for use in large-scale screening applications in the future, subject to its validation in multiple centers.