Bayesian-Optimized Superpixel-GrabCut for Traceable Optic Disc Segmentation
Updated
Updated · arxiv.org · Aug 31
Bayesian-Optimized Superpixel-GrabCut for Traceable Optic Disc Segmentation
1 articles · Updated · arxiv.org · Aug 31
Summary
Researchers have developed a fully traceable segmentation pipeline for optic disc detection in retinal fundus images, matching state-of-the-art performance.
The method combines superpixel decomposition, GrabCut refinement, and Bayesian optimization, achieving a Dice coefficient of 0.9536 on the Drishti-GS dataset.
This transparent approach offers a deterministic, auditable alternative to black-box deep learning models, enhancing clinical trust and facilitating algorithmic debugging.