Fuzzy Measure-Guided Semi-Supervised Breast Cancer Image Segmentation Network

Histopathological breast cancer images often suffer from structural heterogeneity and unclear or complex boundaries. Acquiring pixel-level annotations is costly, limiting the effectiveness and generalizability of traditional segmentation methods. To address these challenges, we propose the Fuzzy Measure-Guided Semi-Supervised Breast Cancer Image Segmentation Network (FuzMGNet). This approach combines fuzzy measures with Convolutional Recurrent Neural Networks (CRNN) and employs Choquet integral-based non-additive feature fusion. A pseudo-labeling guidance mechanism is used to improve boundary delineation. FuzMGNet captures multi-scale contextual information through hierarchical convolutional encoding and spatial modeling using recurrent units. The fuzzy measure dynamically adjusts feature fusion strategies, enhancing the network's adaptability across different images. The Choquet integral strengthens the model's ability to handle complex dependencies, improving segmentation accuracy. Finally, the pseudo-labeling mechanism enables effective training with limited labeled data. Experimental results show that FuzMGNet significantly outperforms traditional deep learning segmentation methods on the MIAS, BreakHis, and BACH datasets.