Weakly Supervised Gland Segmentation with Class Semantic Consistency and Purified Labels Filtration

• DOI: 10.1609/aaai.v39i3.32306
• Publication Date: 2025-04-11T09:47:25Z
• Authors: Siyang Feng; Huadeng Wang; Chu Han; Zhenbing Liu; Hualong Zhang; Rushi Lan; Xipeng Pan
• Keywords: Filtration (mathematics)

Image-level weakly supervised semantic segmentation (WSSS) reduces the dependence on high-quality data annotation, which plays a crucial role in computational pathology. Benefit from ability to localize objects with only binary labels, Class Activation Map (CAM) is widely used method initial pseudo masks. However, due low contrast among different tissues histopathological images, most existing CAM-based methods perform poorly gland segmentation. We retrospect this process and find that class consistency can guide network effectively distinguish confusing pixels generate fine-grained Specifically, for consistency, we propose Consistency Correlation Attention (CCA) encourage focus contribution of features dependencies. For Multi-scale Pyramid Fusion Pooling (MPFP) aggregate coarse-to-fine global information CAMs at multiple spatial resolutions, thus identifying localization. Additionally, introduce Purified Labels Filtration (PLF) strategy during phase mitigate noisy supervision signal improve quality model. Extensive experiments show our achieves new state-of-the-art results three publicly available datasets. Furthermore, demonstrates impressive domain adaptation capability, achieving satisfactory small portion samples when faced unseen data.