Rijul Saurabh Soans¹, Susana T. L. Chung¹; ¹University of California, Berkeley

Amblyopia is a developmental disorder characterized by impaired vision in one eye and is a leading cause of visual impairment in children. Early intervention in life improves the chances of restoring normal visual functions. Evidence exists suggesting that retinal vasculature may differ in individuals with amblyopia, potentially enabling automated retinal image-based detection of the disorder. This study used computer vision to investigate whether there are structural differences in retinal blood vessels between individuals with amblyopia and neurotypical controls. We first performed retinal vessel segmentation using the Spatial Attention-UNet network on OCT fundus images obtained from 22 adults with amblyopia (9 males; 37.68±18.33 years; 11 anisometropic; 11 strabismic) and 40 neurotypical (control) participants (13 males; 23.45±3.46 years). Four features were then quantified from the segmented vasculature images: Vascular Area (VA): number of pixels within segmented vessels; Fractal Dimension (FD): complexity of vascular network by measuring blood vessel density distribution patterns through regression analysis of vessel-block counts across scaled grid sizes; Vascular Skeleton Length (VSL): pixel count of the skeletonized vessel tree; and Vascular Bifurcation Points (VBP): branching complexity of the vessel skeleton indicating morphological richness. Within the amblyopia group, no significant differences were found between amblyopic and fellow eyes of the amblyopes, or between the anisometropic and strabismic groups, for the four vasculature features. Comparing amblyopic eyes to control eyes, significant differences emerged for VA (p=0.01; Cohen’s d=-0.66), FD (p=0.04; d=-0.56), VSL (p=0.006; d=-0.76), but not for VBP (p=0.83). Fellow eyes also differed from control eyes for VA (p=0.03; d=-0.58), FD (p=0.03; d=-0.58), VSL (p=0.02; d=-0.59), but not for VBP (p=0.93). These results indicate significant structural differences in the retinal vasculature between the eyes (both amblyopic and fellow eyes) of amblyopes and neurotypical participants, suggesting that certain structural features of retinal blood vessels might be useful as biomarkers for detecting amblyopia.