Early Detection of Osteoarthritis Stage by Applying Classification Technique to Human Joint imagery

Abstract

Knee Osteoarthritis (OA) is a degenerative joint disease or degenerative arthritis which is the most common chronic condition of joint inflammation coursing various paints such as joint paint, stiffness, swelling, creaking or creating sound, decrease ability to move and bone spur. It is a major cause of disability in older people. The risk of knee OA increases from age 45 and older. Early diagnosis is typically made using X-ray imagery. The work presented in this thesis has proposed the three different mechanisms for knee OA Classification includes: (i) texture based approach, (ii) graph-based approach, and (c) Convolutional Neural Network (CNN) deep learning approach. The fundamental idea is to segment X-ray image so as to obtain the X-ray pixels describing the region of interest (ROI) which were done manually and representing these segmentations using an appropriate representation mechanism to translate an X-ray image into a form that serves to captures key information while remaining compatible with the classification process. By pairing each representation with its OA stage, a classifier can be generated to predict the OA stage according to the nature of a selected representation. The generated classifier can be then used to provide a quick and easy mechanism for labelling the X-ray imagery. As the result of the three different approach, the texture-based approach produce the best result of AUC (Area Under Curve) value of 0.912 in case of OA detection and AUC value of 0.871 in case of OA stage classification. For graph-based approach, the OA detection performed with the AUC value of 0.917 and OA stage classification with AUC value of 0.819 as the best record of graph-based. Finally, the application of CNN produced the best result of OA detection study with AUC value of 0.880 and OA stage classification with AUC value of 0.629. It can be concluded that all the three approach: (i) Texture-Based, (ii) Graph-Based and (iii) CNN-based are well performed on OA detection, while texture-based and graph-based produced the good result for OA stage classification.