SPADESegResNet: Harnessing Spatially-Adaptive Normalization for Breast Cancer Semantic Segmentation

Annotating tissue regions within whole-slide histology images poses a significant challenge for clinical experts and practitioners. In this work, we propose the SPADESegResNet model, which leverages the power of spatially adaptive normalization layers for classifying image pixels into distinct tissue regions. This model effectively segments breast cancer tissue regions while preserving finer details. The framework is assessed both visually and quantitatively using performance metrics such as the Dice score, AUC-ROC, and pixel classification accuracy. The model achieves a Dice score of 0.77 and an AUC-ROC of 0.79 for the identification of tumor regions. Furthermore, the model demonstrates superior performance compared to baseline models, including UNet and its enhanced version, UNet++. This study contributes to the development of robust and interpretable models for semantic segmentation tasks, indicating potential areas for improvement in deep learning architectures. The implementation will be made publicly available upon publication of the paper.