PViTA: Attention-Enhanced Vision Transformer Approach for Analyzing Pupil Dynamics in Asthenopia.

Area of Science:

• Ophthalmology and Computational Vision

Background:

• Increased digital device use and prolonged close work contribute to Asthenopia, necessitating early diagnosis for visual impairment prevention.
• Pupil response to light is a key indicator of visual function and adaptation, varying with task demands and ambient conditions.

Purpose of the Study:

• To quantify pupillary size variations under different light intensities using a novel deep learning model.
• To develop and validate a framework for analyzing pupil dynamics relevant to digital eye strain.

Main Methods:

• A custom dataset of eye images was created under varying lighting conditions.
• The Pupil-Based Vision Transformer with enhanced Attention (PViTA) model was trained on the MOBIUS eye dataset for pupil segmentation.
• PViTA was applied to the custom dataset to analyze pupil diameter changes across light intensities.

Main Results:

• The PViTA model achieved high accuracy (0.9816) and dice value (0.7015) in pupil segmentation during training.
• Pupil diameter significantly decreased with increasing light intensity, showing a constriction pattern.
• The model accurately predicted pupil size changes, decreasing from 65 to 40 pixels across the tested intensity range.

Conclusions:

• The proposed PViTA framework effectively captures pupil dynamics in response to varying light intensities.
• This approach offers a valuable tool for objective assessment of visual adaptation and potential diagnosis of conditions like Asthenopia.