Nested Radially Monotone Polar Occupancy Estimation: Clinically-Grounded Optic Disc and Cup Segmentation for Glaucoma Screening

arXiv cs.CV / 4/13/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper argues that existing deep learning for optic disc (OD) and optic cup (OC) segmentation may violate clinical constraints such as star-convexity and nested OD/OC structure, harming glaucoma diagnostic metrics under cross-dataset shifts.
It introduces NPS-Net (Nested Polar Shape Network), which formulates OD/OC segmentation as nested radially monotone polar occupancy estimation to enforce clinical-valid anatomical structure.
Across seven public fundus datasets, the approach demonstrates strong zero-shot generalization and improved robustness to domain shift.
On RIM-ONE, NPS-Net achieves 100% anatomical validity and improves Cup Dice by 12.8 absolute percentage points while reducing vCDR MAE by more than 56% compared with the best baseline.
On PAPILA, it reports Disc Dice of 0.9438 and Disc HD95 of 2.78 px, with an 83% reduction in HD95 versus the best competing method.

Abstract

Valid segmentation of the optic disc (OD) and optic cup (OC) from fundus photographs is essential for glaucoma screening. Unfortunately, existing deep learning methods do not guarantee clinical validness including star-convexity and nested structure of OD and OC, resulting corruption in diagnostic metric, especially under cross-dataset domain shift. To adress this issue, this paper proposed NPS-Net (Nested Polar Shape Network), the first framework that formulates the OD/OC segmentation as nested radially monotone polar occupancy estimation.This output representation can guarantee the aforementioned clinical validness and achieve high accuracy. Evaluated across seven public datasets, NPS-Net shows strong zero-shot generalization. On RIM-ONE, it maintains 100% anatomical validity and improves Cup Dice by 12.8% absolute over the best baseline, reducing vCDR MAE by over 56%. On PAPILA, it achieves Disc Dice of 0.9438 and Disc HD95 of 2.78 px, an 83% reduction over the best competing method.