Minimal Knee Joint Space Width Detection in Digital X-Ray Images using Deep Learning

This analysis aimed to evaluate the performance of a deep learning-driven method to locate minimal knee joint space width area (KJSW) and see how successfully a deep learning approach can locate mKJSW area from digital x-ray images. In 1848 knee data (digital X-ray images), I used 1752 data (X-ray images) for the study. This proposed method compares the KJSW area in lateral and medial sides, and detect the mKJSW area. Faster RCNN architecture and incorporate VGG -16 with transfer learning are trained to extract the image features to locate the mKJSW area in the digital knee joint X-ray images. The proposed approach located the mKJSW region with a localization accuracy of 98.6 %, indicating the potential use of mKJSW area detection. We have presented the automated deep learning approach to locate the mKJSW area. In addition to that, our proposed deep learning approach trained on both femur and tibial bone region from knee Anterior-Posterior (AP) view digital X-ray images perform better at locating mKJSW area than the reference model. The proposed approach will be used to evaluate knee joint OA and pain assessment.