Deep Learning Based Segmentation Of The Posterior Eye To Study Glaucoma and Age Related Morphological Changes
Loading...
Date
2024-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
The Ohio State University
Abstract
Glaucoma, the leading cause of irreversible blindness in the United States, poses therapeutic challenges with current methods relying on IOP (intraocular pressure) management. However, even after effective lowering of IOP, patients sometimes can still lose vision permanently. Recent studies highlight the potential association between peripapillary sclera (PPS) and lamina cribrosa (LC) biomechanical properties and glaucoma risk. This study aims to explore morphological differences in posterior ocular tissues in order to understand more effective alternatives of focusing treatments for glaucoma.
This study utilized a 50 MHz Vivo 2100 ultrasound imaging system for posterior eye assessment using both healthy and glaucomatous human donor eyes. A portion of the ultrasound images were processed and manually segmented using a custom software. Using this segmented data, a DeepLab V3+ DL-based auto-segmentation tool was employed with a Timm-Efficient Net encoder was trained on for more efficient and streamlined segmentation of the remaining data set. The DL model demonstrated good accuracy in auto-segmentation, surpassing 90% accuracy in relevant ocular tissue layers. Morphological data consisting of the thickness of the retina, sclera, and LC as well as depth between the Bruch’s membrane opening (BMO) and anterior LC were calculated in MatLab. Our results showed a significant difference in thickness between healthy and glaucoma eyes in the retina and age-related thinning of the PPS demonstrating the potential of DL-based auto-segmentation for ocular tissue morphological assessment in glaucoma therapeutics.
Description
Keywords
Learning, Glaucoma, Segmentation, Eye, Deep