Grading of Knee Osteoarthritis Using Convolutional Neural Networks

Knee osteoarthritis (OA) is a disease of the joints and a leading cause of disability among the elderly. If detected at an early stage, its advancement can be slowed and the patient's suffering can be reduced. A new approach involving multiscale convolutional blocks in convolutional neural network (MCBCNN) has been introduced in this paper for automatic classification and grading of knee OA. The proposed model is implemented using pretrained convolutional neural networks (CNNs) and multiscale convolutional filters. Three pretrained CNN models, namely mobileNet2, resNet50 and inceptionNetv3 have been used for the implementation of MCBCNN. Exhaustive performance analysis has been conducted on the three proposed models. The results of knee OA grading delivered by all the three proposed MCBCNNs have been compared. The results show that the performance of MCBCNNs is better than that of the pretrained CNNs. Among the proposed three MCBCNNs, the MCB resNet50 delivers better performance in terms of average accuracy of over 95%, area under curve of nearly 0.9 and F1 score of 0.8.

Automated summary

Knee osteoarthritis is a major cause of disability in the elderly, and early detection can slow its progression and reduce patient suffering. The use of MCBCNN, a new approach involving multiscale convolutional blocks in convolutional neural network, allows for automatic classification and grading of knee osteoarthritis.