Deep Neural Networks for Automatic Classification of Knee Osteoarthritis Severity Based on X-ray Images

Abstract

Knee Osteoarthritis (KOA) is a type of chronic disease that commonly occurs in older, obese citizens and those with a sedentary lifestyle. This disease causes damage to knee cartilage and causes pain so that the patient's activity is reduced. Radiologists classify the KOA severity based on Joint Space Narrowing (JSN) and the presence or absence of osteophytes into five stages from healthy knee (stage 0) to the worst damage (stage 4). We developed a methodology that aims to accelerate the classification of KOA severity based on information obtained from X-ray images and to reduce the subjectivity of radiologists. This paper describes an automated KOA diagnostic model using hyper-parameter Deep Convolutional Neural Networks (DCNN). Based on our experimental result, it shows the accuracy of the proposed method outperforms other KOA severity classification algorithms, which also discussed deep learning, namely 77.24%. This value is the average result of the accuracy of each fold from each stage of the KOA severity level where we use three-folds cross validation as a method of evaluating system performance. Thus computationally, this method is efficient in automatic diagnosis and has the potential to be a clinician application aid to specify the KOA severity.