Applying Texture Analysis and Deep Learning to Assess Photoreceptor Function and Retinal Blood Flow in Adaptive Optics
Date of Award
Summer 7-13-2026
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Biomedical Engineering
First Advisor
Robert Cooper
Second Advisor
Bing Yu
Third Advisor
Richard Povinelli
Abstract
Adaptive optics scanning laser ophthalmoscopy (AOSLO) enables cellular-level assessment of retinal structure and function, but existing methods for quantifying photoreceptor response and vascular dynamics have important limitations. Intensity-based optoretinography (iORG) discards spatial and textural information within the cone aperture during signal extraction, and manual measurement of erythrocyte streak angles for blood velocity estimation is labor-intensive and difficult to scale. This thesis addresses both limitations through two aims. Aim 1 applied grey-level co-occurrence matrix (GLCM)-based texture analysis to develop texture-based optoretinography (tORG), a method for detecting stimulus-evoked changes in cone outer segment (COS) texture. Using AOSLO data from four northern tree shrews, tORG signals showed clear, stimulus-driven changes across contrast, energy, and homogeneity metrics that were distinguishable from control trials and temporally correlated with corresponding iORG responses, supporting tORG as a source of functional information not captured by intensity-based measures alone. Aim 2 developed TemporalResCNN, a ResNet-18 and gated recurrent unit (GRU) based deep learning model, to predict retinal blood flow velocity from standard AO video frames and speckle contrast maps, without requiring specialized XT acquisition. The trained model generated smooth, temporally consistent velocity predictions across eight full inference videos, achieving RMSE values as low as 3.92 px/s and Pearson correlation coefficients as high as 0.68 relative to hand-measured ground truth. Together, these results establish texture-based analysis as a viable, scalable complement to standard AOSLO metrics for assessing both photoreceptor function and retinal vascular dynamics.