Quantitative evaluation of angle-closure parameters in ultrasound biomicroscopy based on dual-module U-Net deep learning technology

Key Points

• This research aims to develop an automated analysis system for assessing angle-closure parameters in PACD.
• Designed a deep learning-based image analysis system for ultrasound biomicroscopy (UBM).
• Constructed a dual-module U-Net architecture for tissue segmentation and scleral spur localization tasks.
• Utilized computer vision algorithms for automatic calculation of multiple angle parameters.
• Achieved high segmentation performance with mean intersection-over-union (IoU) scores of 0.93 for cornea-sclera, 0.92 for iris, and 0.87 for ciliary body.
• Scleral spur localization with an average error of 60.24 microns.
• Consistency of parameter measurements verified by intra-class correlation coefficient (ICC) values exceeding 0.97.

Abstract