Osteo-Doc: KL-Grading of Osteoarthritis Using Deep-Learning

Abstract

Various deep learning frameworks are being proposed for autonomous detection of diseases to contribute towards telemedicine. Moreover, in spite of low doctor to patient ratio, such algorithms aid physicians in tracking the disease with more accuracy. According to WHO, Osteoarthritis has been declared as the most common form of arthritis. Additionally, it is one of the major reasons of physical disability among older age. Different deep learning framework-based approaches exist for evaluation of Knee osteoarthritis but none of them incorporate the feedback or symptoms of the patients. We have proposed a tri-weightage classification model i.e. a hybrid approach for grading osteoarthritis using structural features from X-Ray images, KOOS questionnaire and flexion angle. Moreover, we conducted a comparison of various deep learning model on our dataset and achieved the highest accuracy of 89.29% for RESNET152V2 and INCEPTIONRESNETV2.