FASNet: Feature alignment-based method with digital pathology images in assisted diagnosis medical system

• DOI: 10.1016/j.heliyon.2024.e40350
• Publication Date: 2024-11-13T13:43:46Z
• Authors: Keke He; Jun Zhu; Limiao Li; Fangfang Gou; Jia Wu
• Keywords: Normalization; Digital Pathology; Feature (linguistics); Data pre-processing; Categorical variable

Many important information in medical research and clinical diagnosis are obtained from images. Among them, digital pathology images can provide detailed tissue structure cellular information, which has become the gold standard for tumor diagnosis. With development of neural networks, computer-aided presents identification results various cell nuclei to doctors, facilitates cancerous regions. However, deep learning models require a large amount annotated data. Pathology expensive difficult obtain, insufficient annotation data easily lead biased results. In addition, when current evaluated on an unknown target domain, there errors predicted boundaries. Based this, this study proposes feature alignment-based detail recognition strategy image segmentation (FASNet). It consists preprocessing model network (UNW). The UNW performs instance normalization categorical whitening by inserting semantics-aware modules into encoder decoder, achieves compactness features same class separation different classes. FASNet method identify more efficiently, thus differentiate between classes tissues effectively. experimental show that Dice Similarity Coefficient (DSC) value 0.844. good performance even faced with test does not match distribution training Code: https://github.com/zlf010928/FASNet.git.