Multi-Plane Hybrid Fusion for Robust Knee MRI Diagnosis
Abstract
Automated knee MRI analysis requires deep learning models that are not only accurate but also robust and clinically interpretable across different anatomical planes. In this study, we systematically evaluated four architectures: Baseline MRNet, Mamba, Hybrid, and a proposed Hybrid Fusion framework for ACL and meniscus tear classification using sagittal, coronal, and axial knee MRI from the MRNet dataset. Experiments were conducted using three independent random seeds, and performance was reported as mean ± standard deviation to assess both diagnostic accuracy and training stability. The proposed Hybrid Fusion framework achieved strong and competitive performance across most experimental settings. In ACL classification, the highest single-plane performance was obtained on sagittal MRI with an AUC of 0.976, while the three-plane Hybrid Fusion configuration achieved an AUC of 0.949, demonstrating the ability to integrate complementary anatomical information across multiple MRI planes. In meniscus classification, which remained more challenging across all architectures, the proposed framework achieved competitive performance with AUC values of 0.791 in sagittal and 0.826 in axial MRI. Interestingly, the Mamba-based architecture consistently demonstrated lower mean AUC values and higher variance compared with CNN-based approaches, suggesting that long-range sequential dependency modeling may provide limited benefit in plane-specific knee MRI analysis where diagnostically relevant findings are often localized within adjacent slices and anatomical regions. In contrast, CNN-based representations provided more stable feature learning under limited-data clinical settings. To improve interpretability and clinical reliability, Grad-CAM visualizations were additionally incorporated, demonstrating that the proposed framework focuses on anatomically relevant joint structures during decision making. Overall, the findings suggest that anatomically informed hybrid fusion strategies provide a promising and clinically meaningful direction for robust multi-plane knee MRI analysis.
Keywords
- Knee MRI
- ACL tear detection
- Meniscus tear detection
- Multi-plane learning
- Hybrid fusion
- Deep learning
- Medical Image Analysis
- Mamba
Ethical Statement
References
- Abdullah, R. H., Khattab, R. T., Ahmed, A. R., & Hatif, R. M. (2017). Role of magnetic resonance imaging in evaluation of anterior cruciate ligament injuries. The Egyptian Journal of Hospital Medicine, 69(7), 2897–2905. https://doi.org/10.12816/0042584
- Astuto, B., Flament, I., Namiri, N. K., Shah, R., Bharadwaj, U., Link, T. M., Bucknor, M. D., Pedoia, V., & Majumdar, S. (2021). Automatic deep learning-assisted detection and grading of abnormalities in knee MRI studies. Radiology: Artificial Intelligence, 3(3), e200165. https://doi.org/10.1148/ryai.2021200165
- Azcona, D., McGuinness, K., & Smeaton, A. F. (2020). A comparative study of existing and new deep learning methods for detecting knee injuries using the MRNet dataset. In Proceedings of the 2020 International Conference on Intelligent Data Science Technologies and Applications (IDSTA) (pp. 149–155). IEEE. https://doi.org/10.1109/IDSTA50958.2020.9264258
- Bien, N., Rajpurkar, P., Ball, R. L., Irvin, J., Park, A., Jones, E., Bereket, M., Patel, B. N., Yeom, K. W., Shpanskaya, K., Halabi, S., Zucker, E., Fanton, G., Amanatullah, D. F., Beaulieu, C. F., Riley, G. M., Stewart, R. J., Blankenberg, F. G., Larson, D. B., Jones, R. H., & Lungren, M. P. (2018). Deep learning-assisted diagnosis for knee magnetic resonance imaging: Development and retrospective validation of MRNet. PLoS Medicine, 15(11), e1002699. https://doi.org/10.1371/journal.pmed.1002699
- Dunnhofer, M., Martinel, N., & Micheloni, C. (2022). Deep convolutional feature details for better knee disorder diagnoses in magnetic resonance images. Computerized Medical Imaging and Graphics, 102, 102142. https://doi.org/10.1016/j.compmedimag.2022.102142
- Faruch-Bilfeld, M., Lapègue, F., Chiavassa, H., & Sans, N. (2016). Imaging of meniscus and ligament injuries of the knee. Diagnostic and Interventional Imaging, 97(7–8), 749–765. https://doi.org/10.1016/j.diii.2016.07.003
- Joshi, K., & Suganthi, K. (2024). Anterior cruciate ligament tear detection based on convolutional neural network and generative adversarial neural network. Neural Computing and Applications, 36, 5021–5030. https://doi.org/10.1007/s00521-023-09350-x
- Li, F., Zhai, P., Yang, C., Feng, G., Yang, J., & Yuan, Y. (2023). Automated diagnosis of anterior cruciate ligament via a weighted multi-view network. Frontiers in Bioengineering and Biotechnology, 11, 1268543. https://doi.org/10.3389/fbioe.2023.1268543
Details
Primary Language
English
Subjects
Deep Learning, Artificial Intelligence (Other)
Journal Section
Research Article
Authors
Fatma Harman
*
0000-0002-5358-3516
Türkiye
Publication Date
June 21, 2026
Submission Date
April 9, 2026
Acceptance Date
June 18, 2026
Published in Issue
Year 2026 Number: 10
