融合共享Net的跨模态脑肿瘤分割方法OA北大核心CSTPCD

A cross-modal segmentation framework based on shared Net is proposed to enhance the generalization ability of the brain tumor image segmentation algorithm.The framework consists of three stages,including style conversion,cross-domain training and adaptive discrimination.The Bézier curve is used for the domain variation,and the invisible target domain is simulated based on a variety of images whose grayscales are different from the source domain.A shared Net model based on the lightweight scale attention module is constructed,and multiple styles of gray images are input into the shared Net to learn the weight information of different domains.The optimal segmentation results are selected adaptively by an adaptive discriminator during model inference.Simulation results show that the proposed shared Net algorithm can achieve effective generalization,which is superior to the most advanced methods in segmentation performance and computational efficiency.