生物医学工程研究2025，Vol.44Issue(2)：97-105,9.DOI:10.19529/j.cnki.1672-6278.2025.02.05
多尺度特征融合的胰腺及其肿瘤影像分割算法
Multi-scale feature fusion algorithm for image segmentation of pancreas and its tumors
摘要
Abstract
Addressing the complex structure of pancreas and its tumors,large differences between sections,intra-section location variation,small target areas and fuzzy boundaries,we proposed a multi-scale feature fusion network MSFSTnet for segmentation of pan-creas and its tumors.Firstly,the multi-scale feature selection module was used to extract multi-scale features,and the attention mech-anism was used to automatically assign weights to the features,so that the network paid more attention to the small features such as boundaries.Additionally,a Transformer module was integrated into the bottleneck layer for feature coding to capture long dependen-cies,and fuse local and global features to realize inter-feature interaction and network cross-slice positioning.Experiments results showed that the Dice coefficient,Jaccard coefficient and recall rate of MSFSTnet for pancreas on MSD dataset reached 89.71％,81.90％and 91.97％,and for tumor reached 84.56％,77.39％and 85.48％,respectively.To further validate the network generaliza-tion capability,the MSFSTnet was tested on magnetic resonance image(MRI)datasets from Qilu Hospital of Shandong University Dezhou Hospital,it also achieved good segmentation effect.This study has significant implications for the early diagnosis and treatment of pancreatic cancer.关键词
医学图像分割/多尺度特征/深度学习/Transformer/注意力机制/胰腺分割Key words
Medical image segmentation/Multi-scale features/Deep learning/Transformer/Attention mechanism/Pancreas segmentation分类
基础医学引用本文复制引用
吴树材,王苹苹,许万博,于欣悦,苏童,张盼,徐云峰,魏本征..多尺度特征融合的胰腺及其肿瘤影像分割算法[J].生物医学工程研究,2025,44(2):97-105,9.基金项目
国家自然科学基金项目(62372280,62402297) （62372280,62402297）
山东省自然科学基金项目(2023QF094,2024MF139) （2023QF094,2024MF139）
青岛市科技惠民示范专项(23-2-8-smjk-2-nsh) （23-2-8-smjk-2-nsh）
齐鲁健康与卫生领军人才计划项目 （）
山东省青年科技人才托举工程 （）
山东中医药大学科学研究基金重点项目. （）
