Abstract
Segmentations of retinal optical coherence tomography (OCT) images provide valuable information about each specific retinal layer. However, processing images from degenerative retina remains challenging. This study developed artificial intelligence (AI)-based segmentation to analyze structure changes in sodium iodate (SI)-treated mice. The software is capable of segmenting seven retinal layers and one choroid layer. Analyzing OCT images captured at days post SI-injection (PI) revealed early changes in the retinal pigment epithelium (RPE) layer, with increase in thickness and reduction in reflectance calculated by estimated Attenuation Coefficients (eAC). On the other hand, eAC for outer nuclear layer (ONL) exhibited early and sustained increase after SI treatment. SI induced exponential reduction in ONL thickness with a half-reduction time of about 3 days, indicating progressive photoreceptor degeneration. The extent of degeneration was correlated with ONL eAC level at PI1. Inner retinal layers showed bi-phasic reactions, with initial increases in layer thickness that peaked at around PI3, followed by gradual reduction to lower than baseline levels. In addition, SI also induced transient increases in vitreous particles concentrated around the optic nerve head. Furthermore, there was a gradual reduction of choroid thickness after SI treatment. These results indicate the AI-segmentation tool's usefulness for providing a sensitive and accurate assessment of structure changes in diseased retina and revealed more detailed characterization of SI-induced degeneration in all retinal layers with distinct time courses. Our results also support ONL reflectance changes as an early biomarker for retinal degeneration.