Quantifying the Eye's Response to Light: A Convolutional Neural Network Based Analysis of Pupil Diameter and Palpebral Fissure

Abstract

Purpose : The Ocular Photosensitivity Analyzer (OPA) is an automated test that quantifies a person’s visual photosensitivity threshold (VPT). To gain more information about a subject’s discomfort to light, biometric data of the subject’s eyes can provide objective physiological data. This pilot study will employ a convolutional neural network (CNN) approach to quantify the dynamic pupil and palpebral responses associated with visual photosensitivity.

Methods : The OPA presents a series of light stimuli in an alternating ascending/descending pattern of intensity to calculate a VPT. Throughout the OPA VPT measurement, synchronized, real-time infrared video is recorded of the subject’s eyes (60fps, 6pixel/mm). After the test, the video is analyzed to extract pupil and palpebral fissure area. Frames are segmented with a PSPNet-50 network model, a CNN with custom trained weights implemented through an open-source deep learning-based semantic segmentation library and morphological post-processing (Figure 1). Bilateral pupil and palpebral fissure area are correlated to light intensity of the OPA test.

Results : Four healthy subjects and 3 subjects with TBI underwent OPA VPT test and their videos were post-processed to extract pupil and palpebral fissure area (Figure 1). The pupil and palpebral fissure areas were analyzed by comparing the data collected during the Light-ON (stimulus) or Light-OFF (no stimulus) phases of the OPA. In the Light-OFF phase, a significant difference in pupil area was observed in the left (OS) eye between groups (p = 0.03), as well as in the percent constriction in response to stimuli (Healthy 51.7% versus TBI 28.0%, p = 0.05) (Table 1). Additionally, palpebral fissure area showed significant differences between groups for both the left eye (Light-OFF OS, p = 0.03; Light-ON OS, p=0.04) and right eye (Light-OFF OD, p=0.02; Light-ON OD, p=0.03). These findings suggest a potential indication of increased squinting in TBI patients.

Conclusions : High-resolution video acquisition and CNN segmentation allows for quantification of ocular changes that occur in response to changing light intensities using the OPA. Pupil and palpebral fissure areas can be continuously assessed as variables dependent on varying light intensities. These physiological parameters can provide deeper insight into the factors contributing to visual photosensitivity in TBI patients.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.