[1]严武军,李建昌,叶金霞.轻量化GEC-3DUnet骶髂关节影像自动分割[J].计算机技术与发展,2025,(02):33-40.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0323]
　YAN Wu-jun,LI Jian-chang,YE Jin-xia.Lightweight GEC-3DUnet Based Automatic Segmentation of Sacroiliac Joint Medical Imaging[J].,2025,(02):33-40.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0323]
点击复制

轻量化GEC-3DUnet骶髂关节影像自动分割()

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2025年02期
页码:
33-40
栏目:
媒体计算
出版日期:
2025-02-10

文章信息/Info

Title:
Lightweight GEC-3DUnet Based Automatic Segmentation of Sacroiliac Joint Medical Imaging
文章编号:
1673-629X(2025)02-0033-08
作者:
严武军; 李建昌; 叶金霞
太原师范学院 计算机科学与技术学院,山西 晋中 030600
Author(s):
YAN Wu-jun; LI Jian-chang; YE Jin-xia
School of Computer Science and Technology,Taiyuan Normal University,Jinzhong 030600,China
关键词:
Unet; 骶髂关节; 语义分割; 轻量化; 注意力机制
Keywords:
Unet; sacroiliac joint; semantic segmentation; lightweight; attention mechanism
分类号:
TP399
DOI:
10.20165/j.cnki.ISSN1673-629X.2024.0323
摘要:
强直性脊柱炎( AS) 是一种慢性炎症性风湿病,严重影像人们生活质量,被称为不死得癌症。 骶髂关节炎(Sacroiliitis,SI)是 AS 的病理标志和早期表现,早期的骶髂关节诊断治疗可以很好地预防强直性脊柱炎的发生。 利用深度学习对骶髂关节进行自动分割可以大幅度提高医生的诊断效率,然而目前骶髂关节领域自动分割研究相对匮乏,3D 网络也面临计算瓶颈。 针对以上问题,提出了一种基于轻量化 GEC-3DUnet 骶髂关节分割网络。 首先,通过将 Ghost 模块扩展到 3D 网络以线性运算降低网络的参数量,随后引入轻量化的 Coordinate Attention 以提高网络的分割精度。 在山西白求恩医院提供的数据集中验证实验的准确性。 结果表明:Ghost 模块可以在大幅减少网络参数的情况下保持网络性能,而 Coordinate Attention 可以有效提高分割的准确度。 GEC-3DUnet 为高精度、轻量化的骶髂关节分割提供了解决方案,为骶髂关节的自动分级诊断提供了前提条件,在骶髂关节炎相关研究中具有积极意义。
Abstract:
Ankylosing spondylitis ( AS) is a chronic inflammatory rheumatic disease that severely influences the quality of life,often referred to as an undying cancer. Sacroiliitis ( SI) is a pathological hallmark and early manifestation of AS,and early diagnosis and treatment of sacroiliitis can effectively prevent the onset of ankylosing spondylitis. Utilizing deep learning for automatic sacroiliac joint segmentation can significantly enhance the efficiency of doctors’ diagnoses. However,current research on automatic segmentation in the sacroiliac joint field is relatively scarce, and 3D networks face computational bottlenecks. To address these issues, we propose a lightweight GEC-3DUnet sacroiliac joint segmentation network. Firstly,the network extends the Ghost module to a 3D network to reduce the number of network parameters through linear operations. Subsequently,lightweight Coordinate Attention is introduced to improve the network’s segmentation accuracy. The accuracy of the experiments was validated on data provided by Shanxi Bethune Hospital. The results indicate that the Ghost module can maintain network performance while significantly reducing the number of parameters, and Coordinate Attention can effectively improve segmentation accuracy. GEC - 3DUnet provides a solution for high - precision and lightweight sacroiliac joint segmentation,offering a prerequisite for the automatic grading diagnosis of sacroiliac joints and having positive implications for sacroiliitis-related research.

相似文献/References:

[1]陈博伦,周航,王铁军,等.结合注意力与轴向卷积的荧光图像超分辨率方法[J].计算机技术与发展,2025,(03):26.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0338]
　CHEN Bo-lun,ZHOU Hang,WANG Tie-jun,et al.A Super-resolution Method of Fluorescence Images Combining Attention and Axial Convolution[J].,2025,(02):26.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0338]
[2]史健婷,李雪瑶,陈亦男.基于MGM-Unet的轻量级肾脏图像分割算法[J].计算机技术与发展,2025,(06):42.[doi:10.20165/j.cnki.ISSN1673-629X.2025.0034]
　SHI Jian-ting,LI Xue-yao,CHEN Yi-nan.A Lightweight Kidney Image Segmentation Algorithm Based on MGM-Unet[J].,2025,(02):42.[doi:10.20165/j.cnki.ISSN1673-629X.2025.0034]

更新日期/Last Update: 2025-02-10