机电工程技术2025，Vol.54Issue(24)：14-19,25,7.DOI:10.3969/j.issn.1009-9492.2025.00019
小波集成扩展长短期记忆网络用于盆腔骨骼不完全性骨折分割
Wavelet Integrated Extended Long Short-term Memory Network for Pelvic Insufficiency Fracture Segmentation
摘要
Abstract
For pelvic insufficiency fracture following pelvic bone radiation therapy,existing convolutional neural networks are inadequate in capturing contextual information and detailed features due to the large differences in the size,shape and boundary of fracture areas.To solve these problems,a deep learning model WTF-UxLSTM based on wavelet transform and extended long short-term memory network is proposed.The model introduces mLSTM blocks instead of traditional convolutional blocks.mLSTM enhances the ability to capture image context information by expanding the operation of the memory unit from scalar to matrix,and integrates high-frequency features extracted by wavelet transform into the encoder,thus improving the accuracy of the model in small target fractures and irregular region segmentation.A feature fusion module is designed to enhance the recognition ability of fracture regions through efficient fusion of global and local features,and improve the accuracy of the model's location of fracture regions.The experimental results show that the proposed WTF-UxLSTM model has outstanding performance on several key performance indicators,including Dice coefficient(91.64%),true positive rate(91.43%),specificity(99.93%),average symmetric surface distance(1.265 8 mm),and Hausdorff distance(6.955 7 mm).These indices reflect the good segmentation results of the model.关键词
医学图像分割/扩展长短期记忆网络/小波变换/盆腔骨骼不完全性骨折Key words
medical image segmentation/extended long short-term memory network/wavelet transform/pelvic insufficiency fracture分类
信息技术与安全科学引用本文复制引用
梁淑芬,吴岑,秦传波,张少东,陈诺,肖林..小波集成扩展长短期记忆网络用于盆腔骨骼不完全性骨折分割[J].机电工程技术,2025,54(24):14-19,25,7.基金项目
2024年广东省普通高校重点研究平台项目(2024ZDZX1008) （2024ZDZX1008）
广东省医学科研基金项目(A2019215,B2023100) （A2019215,B2023100）
