Purpose: Objective assessment of visual fatigue remains challenging in contemporary research. This study explored the objective markers of visual fatigue using conventional eye-tracking and an innovative spectacle frame equipped with photodiodes and infrared emitters. Methods: Sixty participants underwent a 30-minute reading task designed to induce visual fatigue. Subjective visual discomfort was assessed using three questionnaires: DESQ (Digital Eye Strain Questionnaire), OSDI (Ocular Surface Disease Index), and SSQVF (Subjective Symptom Questionnaire for Visual Fatigue) before and after the task. Eye movement parameters were recorded using an Eyelink 1000 Plus eye-tracker, while a prototype spectacle frame with nine photodiodes measured periocular region changes. A time series transformer model was trained to classify visual fatigue states using signals from both the Eyelink tracker and the photodiode-equipped frame. Results: Subjective visual discomfort increased significantly across all questionnaires following the reading task. Several eye movement parameters showed significant changes: saccade amplitude and velocity increased, pupil size decreased, and blink duration increased. Correlations between objective measures and subjective reports were generally limited, with only fixation stability metrics (Bivariate Contour Ellipse Area and isoline area) showing significant correlations with SSQVF scores after correction for multiple comparisons. The time series transformer model achieved superior performance using photodiode data (82.5% user-level accuracy) compared with eye-tracking data (73.1% accuracy). Conclusions: Although traditional eye-tracking metrics showed limited correlation with subjective fatigue, with the exception of fixation stability measures, the photodiode-equipped wearable device demonstrated promising potential for objective visual fatigue detection. The superior performance of the photodiode signals suggests this technology could provide a practical alternative to laboratory-based eye-tracking for real-world fatigue monitoring applications.
Guégan et al. (Thu,) studied this question.