[1]彭志博,陈勇,崔艳荣.基于YOLOv8m的改进腕部X光片骨折检测算法[J].中国医学物理学杂志,2025,42(4):542-549.[doi:10.3969/j.issn.1005-202X.2025.04.017]
　PENG Zhibo,CHEN Yong,CUI Yanrong.Fracture detection in wrist X-ray image using an improved algorithm based on YOLOv8m[J].Chinese Journal of Medical Physics,2025,42(4):542-549.[doi:10.3969/j.issn.1005-202X.2025.04.017]
点击复制

基于YOLOv8m的改进腕部X光片骨折检测算法()
分享到：

《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]

卷:
42
期数:
2025年第4期
页码:
542-549
栏目:
医学人工智能
出版日期:
2025-04-20

文章信息/Info

Title:
Fracture detection in wrist X-ray image using an improved algorithm based on YOLOv8m
文章编号:
1005-202X（2025）04-0542-08
作者:
彭志博; 陈勇; 崔艳荣
长江大学计算机科学学院，湖北 荆州 434000
Author(s):
PENG Zhibo;  CHEN Yong;  CUI Yanrong
School of Computer Science, Yangtze University, Jingzhou 434000, China
关键词:
X射线; 骨折检测; 深度学习; YOLOv8
Keywords:
X-ray;  fracture detection;  deep learning;  YOLOv8
分类号:
R318；TP391
DOI:
10.3969/j.issn.1005-202X.2025.04.017
文献标志码:
A
摘要:
目前腕部X光片的骨折检测存在误诊率高、医疗资源不足等问题。为了辅助医生进行骨折诊断，提出了一种基于YOLOv8m的X光片骨折检测方法。首先引入可分离大核注意力机制来提取重要特征信息，抑制不显著特征信息；然后将残差块融入注意力机制，增强注意力机制的作用，增加模型的泛化能力；最后将可切换空洞卷积与C2f模块结合，增加模型的感受野，捕捉不同尺寸的特征信息。实验结果表明，与先进的YOLOv8l改进模型相比，本文模型mAP50提高了1.3%，由于使用了规格更小的YOLOv8m为基础模型，参数量降低了14.3%，浮点运算次数降低了42.7%。此模型能够辅助放射科医生进行腕部X光片的骨折诊断。
Abstract:
Currently, the fracture detection in wrist X-ray image has high misdiagnosis rates and faces the challenge ofinadequate medical resources. To assist doctors in fracture diagnosis, an improved approach based on YOLOv8m for fracturedetection in wrist X-ray image is proposed: (1) a large separable kernel attention mechanism is introduced to extract crucialfeature information while suppressing insignificant ones; (2) residual block is integrated into the attention mechanism toenhance its effectiveness and the model’s generalization ability; (3) switchable atrous convolution is combined with the C2fmodule to expand the model’s receptive field, enabling it to capture multi-scale feature information. Experimental resultsdemonstrate that compared with the improved model based on the advanced YOLOv8l, the proposed approach achieves a1.3% increase in mAP50. Notably, by adopting the more compact YOLOv8m model as the basic model, parameter count isreduced by 14.3%, and the floating-point operations per second is lowered by 42.7%. The proposed model can effectively aidradiologists in detecting fractures in wrist X-ray image.

相似文献/References:

[1]王明,李辉.同步辐射X射线的医学应用[J].中国医学物理学杂志,2016,33(12):1253.[doi:10.3969/j.issn.1005-202X.2016.12.016]
　[J].Chinese Journal of Medical Physics,2016,33(4):1253.[doi:10.3969/j.issn.1005-202X.2016.12.016]
[2]骆荣辉,吴朝,田扬超. 基于投影的X射线光栅相衬成像系统的仿真模拟[J].中国医学物理学杂志,2017,34(11):1124.[doi:DOI:10.3969/j.issn.1005-202X.2017.11.009]
　LUO Ronghui,WU Zhao,TIAN Yangchao. Numerical simulation of X-ray grating phase contrast imaging system based on projection[J].Chinese Journal of Medical Physics,2017,34(4):1124.[doi:DOI:10.3969/j.issn.1005-202X.2017.11.009]
[3]张夏静,夏天,何培忠.微焦点类同轴X射线相衬成像实验及相位恢复重建[J].中国医学物理学杂志,2018,35(9):1050.[doi:10.3969/j.issn.1005-202X.2018.09.011]
　ZHANG Xiajing,XIA Tian,HE Peizhong.Micro-focus in-line X-ray phase contrast imaging experiment and phase recovery reconstruction[J].Chinese Journal of Medical Physics,2018,35(4):1050.[doi:10.3969/j.issn.1005-202X.2018.09.011]
[4]方春锋,徐伟,宋圆源,等.两台医科达直线加速器匹配后束流特性比较[J].中国医学物理学杂志,2020,37(1):1.[doi:DOI:10.3969/j.issn.1005-202X.2020.01.001]
　FANG Chunfeng,XU Wei,SONG Yuanyuan,et al.Comparison of beam characteristics after beam matching in two Elekta linacs[J].Chinese Journal of Medical Physics,2020,37(4):1.[doi:DOI:10.3969/j.issn.1005-202X.2020.01.001]

备注/Memo

备注/Memo:
【收稿日期】2024-12-20【基金项目】国家自然科学基金（62077018）【作者简介】彭志博，硕士研究生，研究方向：深度学习、计算机视觉，E-mail: 2023720815@yangtzeu.edu.cn【通信作者】陈勇，高级工程师，硕士生导师，研究方向：WEB 信息处理、人工智能，E-mail: 285527563@qq.com
更新日期/Last Update: 2025-04-30