Not All Pixels Are Equal: Pixel-wise Meta-Learning for Medical Segmentation with Noisy Labels

ArXi:2511.18894v5 Announce Type: replace-cross Medical image segmentation is crucial for clinical applications, but it is frequently disrupted by noisy annotations and ambiguous anatomical boundaries, limiting its application in real-world scenarios. Existing methods often directly adapt noisy label learning techniques designed for instance classification, overlooking the pixel-wise heterogeneity in medical segmentation with its spatially and anatomically varying difficulties.