| 摘 要: 针对目前膝关节骨折的智能诊断及分析方法存在检测能力不足和识别精度不高的情况,提出一种基于ResNet18融合MSC(Median Enhanced Spatial and Channel Attention Block)注意力机制的识别方法,通过引入MSC注意力模块对ResNet18进行改进和优化,以提升骨折识别的准确率、泛化能力和鲁棒性。实验结果表明,改进后的模型在准确率方面优于当前3种主流注意力机制ECA-Net(Efficient Channel Attention)、Coordinate Attention和Efficient Multi-Scale Attention,训练集识别准确率分别提升了22.9%,21.0%,10.1%。与人工阅片比较,改进模型的诊断时间大幅缩短。此外,模型在测试集上平均准确率达93.0%,验证了本文方法的可行性和有效性。 |
| 关键词: 膝关节骨折 深度学习 卷积神经网络 注意力机制 |
|
中图分类号:
文献标识码: A
|
| 基金项目: 基于有限元分析的交叉韧带生物力学特性与膝骨关节炎致病机理的研究(黔科合基础-ZK[2023]一般052) |
|
| Intelligent Diagnosis of Knee Joint Fracture in Medical Imaging Based on ResNet18 and MSC Attention Mechanism |
|
LIU Yongxing1, WU Peng1, LIU Wenke2
|
(1.State Key Laboratory of Public Big Data Jointly Built by the Provincial and Ministerial Authorities, Guizhou University, Guiyang 550025, China;
2.The Department of Sports Medicine, The Fourth People’s Hospital of Guiyang, Guiyang 550002, China)
544073223@qq.com; pengwu@gzu.edu.cn; Dr.Liu4321@outlook.com
|
| Abstract: To address the limitations in current intelligent diagnosis and analysis methods for knee joint fractures,such as insufficient detection capability and low recognition accuracy, this study proposes a recognition method combining ResNet18 with an MSC(Median-enhanced Spatial and Channel Attention Block)attention mechanism. By introducing the MSC attention module to enhance and optimize ResNet18, the accuracy, generalization ability, and robustness of fracture recognition are improved. Experimental results demonstrate that the improved model outperforms three mainstream attention mechanisms—ECA-Net(Efficient Channel Attention), Coordinate Attention, and Efficient Mult-i Scale Attention—in recognition accuracy. On the training set, accuracy rates increased by 22.9% , 21.0% , and
10.1% , respectively, compared to these models. Furthermore, diagnosis time was significantly reduced relative to manual film reading. The model achieves an average accuracy rate of 93.0% on the test set, validating the feasibility and effectiveness of the proposed method. |
| Keywords: knee joint fracture deep learning convolutional neural network attention mechanism |
