A dual-prototype morphological topology evolution network for clinically oriented brain tumor MRI segmentation.

Automatic segmentation of brain tumor MRI images has important clinical significance for lesion localization, preoperative planning, and therapeutic efficacy assessment. To address the challenges of blurred boundaries, irregular morphology, and similar grayscale appearances between brain tumor regions and surrounding brain tissues, this study proposes a dual-prototype morphological evolution network. The proposed method adopts SegFormer as the backbone network and designs an adaptive dual-prototype representation module to enhance category discrimination through background and tumor prototypes. Meanwhile, a morphological evolution attention module is introduced to improve the model's perception of irregular boundaries and local structures through differentiable soft morphological operations. Experimental results show that the proposed method achieves an mIoU of 0.8479, an mDice of 0.8937, an mRecall of 0.9141, and an HD95 of 23.47 on the public brain tumor dataset. On the clinical brain tumor dataset, the corresponding values reach 0.7937, 0.8422, 0.8673, and 29.86, respectively. These results validate the effectiveness of the proposed method in brain tumor region recognition, boundary localization, and clinical auxiliary segmentation.