Pancreas segmentation network based on multi-scale feature fusion and sliding window attention
LIU Yiqi
XIN Guojiang
YI Xiaolei
LI Xuhui
LIANG Hao
WU Yingjie
Abstract：To address the problem of blurred boundaries and low segmentation accuracy in pancreas segmentation,we proposed a pancreatic segmentation network MSW-HRNet.Firstly,by integrating depthwise separable convolution and spatial attention mecha-nisms,a multi-scale upsample block(MUB)was designed to restore the detailed information during the multi-scale upsampling process,enhance the segmentation ability for small-sized target regions.Then,the sliding-window attention block(Swin-Block)was fused to sense the global context information across scales,improve the discrimination ability of the model on lesion tissue and complex background,and enhance the performance of pancreatic boundary segmentation under complex structures.Experimental results demon-strated that the Dice coefficient of this method reached 82.11%on the NIH public dataset and 86.93%on the private pancreatitis data-set,outperforming mainstream segmentation models,confirming its superiority and practicality in handling complex pathological mor-phologies.
Keywords：Pancreatic segmentationAttention mechanismDepthwise separable convolutionHRNet
Publication Date：2025-12-30
Online Publishing Date：2026-09-11(First online date of this platform, not the publication date of the document)
Pages：8( 371-378 )