Binary Classification of MRI Images of Osteoarthritis with a Modified R3D_18 Model

• Thalles C. Fontainha CEFET/RJ
• Felipe da R. Henriques CEFET/RJ
• Amaro A. Lima CEFET/RJ
• Gabriel M. Araujo CEFET/RJ
• Ricardo de S. Tesch UNIFASE

DOI: https://doi.org/10.5753/sbcas.2025.7629

Abstract

This work proposes a 3D convolutional neural network model derived from the R3D_18 architecture to classify magnetic resonance images (MRI) of the knee, differentiating normal and abnormal images associated with osteoarthritis. Using the OAI-MRI-3DDESS dataset, the model was trained with cross-validation and oversampling to compensate for class imbalance. The approach aims to improve diagnostic accuracy and reduce professionals’ workload by exploring the volumetric information of the images. The experiments indicated robust performance, with accuracy higher than 86% and AUC close to 0.92, showing the method’s potential for clinical applications.

References

Al Turkestani, N., Li, T., Bianchi, J., Gurgel, M., Prieto, J., Shah, H., Benavides, E., Soki, F., Mishina, Y., and Fontana, M. (2024). A comprehensive patient-specific prediction model for temporomandibular joint osteoarthritis progression. Proc. Natl. Acad. Sci. (PNAS), 121(8):e2306132121.

Berrimi, M. (2021). OAI-MRI-3DDESS Dataset. Acesso em: nov. 2024. Disponível em: [link]. Dataset público de ressonância magnética 3D do joelho.

Goodfellow, I., Bengio, Y., and Courville, A. (2016). Convolutional networks. MIT press Cambridge.

Guida, C., Zhang, M., and Shan, J. (2021). Knee osteoarthritis classification using 3D CNN and MRI. Applied Sciences, 11(11):5196.

Hara, K., Kataoka, H., and Satoh, Y. (2018). Can spatiotemporal 3D CNNS retrace the history of 2D CNNs and ImageNet? In Proc. 2018 IEEE/CVF Conf. CVPR, pages 6546–6555, Salt Lake City, USA.

Hosseini-Asl, E., Keynton, R., and El-Baz, A. (2016). Alzheimer’s disease diagnostics by adaptation of 3D convolutional network. In Proc. 2016 IEEE Intl. Conf. Img. Process. (ICIP), pages 126–130, Phoenix, USA.

Kohavi, R. (1995). A study of cross-validation and bootstrap for accuracy estimation and model selection. In 14th Intl. Joint Conf. AI - IJCAI 1995.

Redmon, J., Divvala, S., Girshick, R., and Farhadi, A. (2016). You only look once. In Proc. 2016 IEEE/CVF Conf. CVPR, pages 779–788, Las Vegas, USA.

Ronneberger, O., Fischer, P., and Brox, T. (2015). U-net: Convolutional networks for biomedical image segmentation. In Proc. 18th Intl. Conf. Med. Img. Comp. & Comp.-Asst. Interv. – MICCAI 2015, pages 234–241, Munich, Germany.

Siouras, A., Moustakidis, S., Giannakidis, A., Chalatsis, G., Liampas, I., Vlychou, M., Hantes, M., Tasoulis, S., and Tsaopoulos, D. (2022). Knee injury detection using deep learning on mri studies: a systematic review. Diagnostics, 12(2):537.

Zhong, J., Yao, Y., Khan, S., Xiao, F., Cahill, D. G., Griffith, J. F., and Chen, W. (2022). Knee Osteoarthritis: Automatic Grading with Deep Learning. In Proc. 2022 ISMRM & ISMRT Annu. Mtg. Expo.