Multi-perspective feature reorganization network for medical image segmentation

Medical image segmentation based on encoder–decoder architectures often suffers from insufficient cross-scale semantic fusion and loss of contextual information, which limits segmentation accuracy in complex scenarios. To address these issues, this paper proposes a Multi-Perspective Feature Reorganization Network (MPFRNet) to enhance global context modeling and improve feature interaction across different layers. Specifically, a Multi-layer Large-kernel Convolution Module (MLCM) is introduced to capture hierarchical contextual information, while a Global Context Attention (GCA) mechanism is designed to strengthen feature alignment between encoder and decoder. In addition, a Deep Feature Enhanced Module (DFEM) and a Multi-perspective Feature Reorganization Module (MFRM) are employed to enhance deep semantic representation and mitigate information loss during decoding. Extensive experiments conducted on three public datasets (DRIVE, TNBC, and MoNuSeg) demonstrate that the proposed method achieves consistent improvements over several representative segmentation models in terms of MIoU, Dice, and Accuracy. These results indicate that MPFRNet effectively improves segmentation performance by enhancing semantic consistency and contextual representation, providing a reliable and efficient solution for medical image segmentation tasks.