Pancreas segmentation based on large convolution kernel and feature fusion

Accurate segmentation of the pancreas from CT images is important for the adjuvant treatment of pancreatic diseases. However, due to the complex shape of the pancreas, its small size, and the low contrast with surrounding tissues, achieving high-precision automated segmentation remains highly challenging. This paper proposes a pancreas segmentation method based on large convolution kernels and feature fusion. First, a Dilated Large Kernel (DLK) encoder is designed to achieve a receptive field similar to that of large convolution kernels without significantly increasing computational complexity. Second, a Feature Mix Module (FMM) is introduced to fuse shallow and deep features, enabling feature interaction between upper and lower layers. Finally, a DWLU decoder is developed to preserve local feature extraction capabilities while improving computational efficiency. In order to validate the effectiveness of the network, it was validated on the NIH dataset, and the experimental results showed that DSC, Jaccard and Recall achieved 85.6%, 74.9% and 87.1% respectively on the NIH dataset, which provided an effective solution for the pancreas segmentation task.