| 摘 要: 针对膝关节软骨损伤五分类任务中,医学影像数据获取受限,训练样本匮乏,导致模型特征表征能力不足的问题,提出一种基于层次化影像特征的膝关节软骨损伤分类算法。通过构建包含矢状面、冠状面和横断面的多模态磁共振成像(MRI)数据集,以解决单平面医学影像特征的局限性。此外,首次将 YOLOv11模型应用于膝关节软骨损伤诊断,通过纹理特征与语义特征的跨层级融合,实现层次化特征提取,增强模型对复杂病灶区域的特征表征能力,并基于特征金字塔网络的多尺度融合机制,增强模型鲁棒性。实验结果表明,所提模型的分类精度达到99.83%,相较于传统方法显著提升了分类性能。 |
| 关键词: 膝关节软骨损伤 多级分类 YOLOv11 多模态特征 磁共振成像 |
|
中图分类号:
文献标识码: A
|
| 基金项目: 辽宁省教育厅高校基本科研项目资助(LJ212413631021);辽宁省自然科学基金面上项目(2025-MS-311) |
|
| Hierarchical Image Feature-Based Classification Algorithm for Knee Cartilage Injury |
|
YANG Yating, ZHANG Lirong, YU Hang, WANG Yiqiao, ZHANG Hanye, LI Jiayao
|
( School of Digital Arts & Design, Dalian Neusoft University of Information, Dalian 116023, China)
1095543422@qq.com; zhanglirong1997@163.com; yuhang_ys@neusoft.edu.cn; nnppy1@qq.com; 3267509398@qq.com; 108325487@qq.com
|
| Abstract: Aiming at the problem of insufficient feature representation capability of models caused by limited acquisition of medical imaging data and scarce training samples in the five-class classification task of knee cartilage injury, this study proposes a knee cartilage injury classification algorithm based on hierarchical imaging features. A multi-modal Magnetic Resonance Imaging (MRI) dataset encompassing sagittal, coronal, and transverse planes is constructed to address the limitations of single-plane medical imaging features. Furthermore, the YOLOv11 model is applied for the first time in knee cartilage injury diagnosis. Hierarchical feature extraction is achieved through cross_x005f_x0002_level fusion of texture features and semantic features, enhancing the model’s capability to represent features of complex lesion regions. Additionally, the multi-scale fusion mechanism based on the feature pyramid network is employed to improve model robustness. Experimental results demonstrate that the proposed model achieves a classification accuracy of 99.83% , significantly outperforming traditional methods in classification performance. |
| Keywords: knee cartilage injury multilevel classification YOLOv11 multimodal features magnetic resonance imaging |
