Research on Precision Diagnosis and Treatment Decision of Common Eye Diseases Based on Artificial Intelligence

In this study, we developed a two-module system for the evaluation of congenital ptosis in children based on smartphone video recordings. The first module performs morphological measurements of eyelid parameters (e.g., palpebral fissure height, margin reflex distance) extracted from blink videos. The accuracy of this module was assessed through agreement analysis between automated measurements and manual measurements performed by clinicians.

The second module focuses on levator function grading. The automated grading results were compared against clinical assessments by senior oculoplastic specialists, which served as the gold standard, to evaluate the model's accuracy. Additionally, we compared the model's performance to that of junior ophthalmology residents to assess the system's clinical assistive value in real-world scenarios.

Based on these modules, we developed a smartphone-based application for remote ptosis assessment. A multi-center clinical validation study was designed to evaluate the feasibility, consistency, and generalizability of the system in different clinical settings.

Inclusion criteria

• participants aged ≤18 years with compliance during video recording. Congenital ptosis diagnosis mandated documented symptom onset within the first year of life and bilateral or unilateral MRD1 of < 2.5 mm

Exclusion criteria

• acquired ptosis, myasthenia gravis, Marcus Gunn syndrome, eyelid deformities, or prior ophthalmic surgery