Machine-Learning-Based Automated Schlemm's Canal Volumetric Segmentation for Optical Coherence Tomography.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computational Biology

Background:

• Volumetric segmentation of Schlemm's Canal (SC) in optical coherence tomography (OCT) is crucial for studying glaucoma and trabecular outflow pathways.
• Current manual segmentation methods are time-consuming, hindering in vivo research.

Purpose of the Study:

• To develop an automated tool, Schlemm's Canal-Localization and Semantic Segmentation (SC-LSS), for efficient SC segmentation in mouse eyes using visible-light OCT (vis-OCT).
• To validate the accuracy and utility of SC-LSS in capturing physiological changes in SC size.

Main Methods:

• Developed SC-LSS, a two-stage automated segmentation tool for SC in vis-OCT images.
• Trained and validated the model using 324 B-scans from 16 mouse eyes.
• Evaluated model accuracy against manual expert segmentation and assessed SC size changes under varying intraocular pressure and pilocarpine administration.

Main Results:

• SC-LSS achieved a Dice coefficient of 0.70 ± 0.20, comparable to expert inter-grader variability (0.73 ± 0.18).
• SC-LSS accurately detected a 51.5% decrease in SC size with increased intraocular pressure (20 mmHg vs. 5 mmHg).
• SC-LSS identified a 20.1% increase in SC size after pilocarpine administration.

Conclusions:

• SC-LSS provides accurate and automated volumetric segmentation of Schlemm's Canal in in vivo mouse vis-OCT images.
• This tool is expected to significantly accelerate research into the trabecular outflow pathways and glaucoma.
• SC-LSS demonstrates potential for quantitative analysis of SC dynamics in response to physiological stimuli.