Cross-Layer Connection SegFormer Attention U-Net for Efficient TRUS Image Segmentation

IF 3 4区计算机科学 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC International Journal of Imaging Systems and Technology Pub Date : 2024-09-24 DOI:10.1002/ima.23178

Yongtao Shi, Wei Du, Chao Gao, Xinzhi Li

{"title":"Cross-Layer Connection SegFormer Attention U-Net for Efficient TRUS Image Segmentation","authors":"Yongtao Shi, Wei Du, Chao Gao, Xinzhi Li","doi":"10.1002/ima.23178","DOIUrl":null,"url":null,"abstract":"<div>\n \n <p>Accurately and rapidly segmenting the prostate in transrectal ultrasound (TRUS) images remains challenging due to the complex semantic information in ultrasound images. The paper discusses a cross-layer connection with SegFormer attention U-Net for efficient TRUS image segmentation. The SegFormer framework is enhanced by reducing model parameters and complexity without sacrificing accuracy. We introduce layer-skipping connections for precise positioning and combine local context with global dependency for superior feature recognition. The decoder is improved with Multi-layer Perceptual Convolutional Block Attention Module (MCBAM) for better upsampling and reduced information loss, leading to increased accuracy. The experimental results show that compared with classic or popular deep learning methods, this method has better segmentation performance, with the dice similarity coefficient (DSC) of 97.55% and the intersection over union (IoU) of 95.23%. This approach balances encoder efficiency, multi-layer information flow, and parameter reduction.</p>\n </div>","PeriodicalId":14027,"journal":{"name":"International Journal of Imaging Systems and Technology","volume":"34 5","pages":""},"PeriodicalIF":3.0000,"publicationDate":"2024-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"International Journal of Imaging Systems and Technology","FirstCategoryId":"94","ListUrlMain":"https://onlinelibrary.wiley.com/doi/10.1002/ima.23178","RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"ENGINEERING, ELECTRICAL & ELECTRONIC","Score":null,"Total":0}

引用次数: 0

Abstract

Accurately and rapidly segmenting the prostate in transrectal ultrasound (TRUS) images remains challenging due to the complex semantic information in ultrasound images. The paper discusses a cross-layer connection with SegFormer attention U-Net for efficient TRUS image segmentation. The SegFormer framework is enhanced by reducing model parameters and complexity without sacrificing accuracy. We introduce layer-skipping connections for precise positioning and combine local context with global dependency for superior feature recognition. The decoder is improved with Multi-layer Perceptual Convolutional Block Attention Module (MCBAM) for better upsampling and reduced information loss, leading to increased accuracy. The experimental results show that compared with classic or popular deep learning methods, this method has better segmentation performance, with the dice similarity coefficient (DSC) of 97.55% and the intersection over union (IoU) of 95.23%. This approach balances encoder efficiency, multi-layer information flow, and parameter reduction.

查看原文

微信好友朋友圈 QQ好友复制链接

本刊更多论文

跨层连接 SegFormer 关注 U-Net 实现高效 TRUS 图像分割

由于超声图像中的语义信息非常复杂，因此准确、快速地分割经直肠超声（TRUS）图像中的前列腺仍是一项挑战。本文讨论了跨层连接 SegFormer 注意力 U-Net，以实现高效 TRUS 图像分割。SegFormer 框架通过降低模型参数和复杂度而不牺牲准确性得到了增强。我们引入了层跳连接以实现精确定位，并将局部上下文与全局依赖性相结合，从而实现卓越的特征识别。解码器采用多层感知卷积块注意力模块（MCBAM）进行改进，以实现更好的上采样并减少信息丢失，从而提高准确性。实验结果表明，与经典或流行的深度学习方法相比，该方法具有更好的分割性能，骰子相似系数（DSC）为 97.55%，交集大于联合（IoU）为 95.23%。这种方法兼顾了编码器效率、多层信息流和参数缩减。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文去求助

来源期刊

International Journal of Imaging Systems and Technology 工程技术-成像科学与照相技术

CiteScore

6.90

自引率

6.10%

发文量

138

审稿时长

3 months

期刊介绍： The International Journal of Imaging Systems and Technology (IMA) is a forum for the exchange of ideas and results relevant to imaging systems, including imaging physics and informatics. The journal covers all imaging modalities in humans and animals. IMA accepts technically sound and scientifically rigorous research in the interdisciplinary field of imaging, including relevant algorithmic research and hardware and software development, and their applications relevant to medical research. The journal provides a platform to publish original research in structural and functional imaging. The journal is also open to imaging studies of the human body and on animals that describe novel diagnostic imaging and analyses methods. Technical, theoretical, and clinical research in both normal and clinical populations is encouraged. Submissions describing methods, software, databases, replication studies as well as negative results are also considered. The scope of the journal includes, but is not limited to, the following in the context of biomedical research: Imaging and neuro-imaging modalities: structural MRI, functional MRI, PET, SPECT, CT, ultrasound, EEG, MEG, NIRS etc.; Neuromodulation and brain stimulation techniques such as TMS and tDCS; Software and hardware for imaging, especially related to human and animal health; Image segmentation in normal and clinical populations; Pattern analysis and classification using machine learning techniques; Computational modeling and analysis; Brain connectivity and connectomics; Systems-level characterization of brain function; Neural networks and neurorobotics; Computer vision, based on human/animal physiology; Brain-computer interface (BCI) technology; Big data, databasing and data mining.