Automatic Localization of Knee Joint Area in Plain Radiographs using Adaptive Faster RCNN

Knee joint Osteoarthritis (OA) is a very important reason for the functional limitations and physical illness seen in older people, typically diagnosed from the radiographic assessment. Practitioner evaluation is highly dependent on the experience. Computer-assisted machine learning techniques could enhance the accuracy level of knee joint OA assessment. The initial stage of arthritic knee joint analysis is to locate the knee joint space area region of interest (ROI) automatically in X-ray images. This paper introduces a model for automatic location of knee joint OA based on deep learning approach. Robust and efficient localization of the knee joint task remains a challenge and existing works do not provide satisfactory results. Our aim is to design an effective deep learning driven method to overcome this problem. We trained an adapted deep neural network, namely a faster region convolutional neural network (RCNN) to attain the main objective which is the extraction of the intraarticular knee joint space region from the X-ray images. Results of the proposed improved Faster RCNN based system shows that we obtain a mean average precision of nearly 0.971 indicating the potential use of automatic detection in OA. This proposed work delivers, the best performance on knee joint ROI detection from X-ray images and can be used to quantify the knee joint OA.