Fundus-Based RNFL Quadrant Map Generation Using Depth Estimation

Evaluating the optic nerve head (ONH) and retinal nerve fiber layer (RNFL) is crucial for glaucoma. Though optical coherence tomography (OCT) offers advanced RNFL imaging, its cost limits accessibility. However, fundus images present a valuable alternative, with its availability and lower cost. This study presents a modality for glaucoma assessment using fundus images, i.e., employing depth estimation to generate RNFL thickness maps and delineate TSNIT graphs to create RNFL quadrants. Through this approach, the overall results consist of a prediction accuracy of red, yellow, green and white at 92.32%, 87.56%, 73.01%, and 89.94%, respectively. The micro F1-score is 71.43%. Analysis of the generated RNFL quadrant in the nasal region reveals a dominance of green, suggesting minimal thinning as fewer glaucoma cases occur in this area. Conversely, red dominates after green in superior and inferior quadrants, aligning with known trends of RNFL thinning and higher glaucoma prevalence. Findings highlight the potential of fundus images as a valuable resource for assessing RNFL thickness in glaucoma, offering a promising avenue for efficient and resource-conscious clinical applications.