Abstract
Purpose :
The retinal ganglion cell (RGC) layer integrates and transmits stimuli from photoreceptors to the central nervous system. Loss of RGCs is a hallmark of glaucoma and other retinopathies and neuropathies, and quantifying changes in this cell population is important for understanding disease severity and treatment efficacy in humans and animal models.
Artificial intelligence (AI) has demonstrated utility in automated quantitation of whole slide images (WSI). Specifically, applying AI to the interpretation of hematoxylin and eosin (H&E) stained tissue sections is expanding in both clinical and research settings.
H&E is a standard microscopy stain, easily implemented, highly reproducible, cost-effective and able to delineate cellular details in healthy or diseased tissue; however, subtle microscopic findings often require time consuming assessments, and may be overlooked, even by experienced pathologists. Likewise, it is challenging for the pathologist to manually count cells accurately and reproducibly.
AI presents a tool to overcome these limitations and expand utility and application of H&E stained WSI, which we have applied to automate detection of the macula and RGCs at the level of the fovea in 4um thick cross sections of nonhuman primate globes.
Methods :
AI-enabled image analysis software (Visiopharm, Copenhagen, Denmark) was applied to WSI of 19 H&E-stained African green monkeys’ eyes from which 13 (~70%) were used for training, 3 (~15%) for validation, and 3 (~15%) for testing. We designed an image analysis and AI-driven workflow to find the macula in each eye vs. the peripheral retina; and then quantitate the number of RGC in that specific region of interest. Results were validated by comparing pathologist-guided manual annotations to the AI results to calculate the accuracy, sensitivity, and specificity.
Results :
This method distinguishes macula from peripheral retina, and defines a numerical value for macula area, number of RGC within the macula, and the total area occupied by RGCs. Our method shows excellent results: 89.47% accuracy on finding the macula and 97.25% accuracy for RGC counts.
Conclusions :
This study provides proof of concept for automated detection and quantitation of macula and RGC in H&E stained WSI. Our approach promises to enhance accuracy and improve histopathology workflows.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.