SG-Transunet: A segmentation-guided Transformer U-Net model for KRAS gene mutation status identification in colorectal cancer

Yulan Ma, Yuzhu Guo, Weigang Cui, Jingyu Liu, Yang Li, Yingsen Wang, Yan Qiang

Abstract: Highlights•We utilize the segmentation task to facilitate classification, which helps the model focus on lesion regions.•An aggregation attention block consolidates shallow, deep, and spatial-frequency features, promoting a more comprehensive feature representation.•A mutual-constrained loss function is designed to simultaneously constrain the segmentation mask acquisition and gene status discrimination process.