A CNN Based Approach for the Prediction of Knee Osteoarthritis

Abstract

Radiology takes place a crucial part in various medical applications like methods used for earlier diagnosis, observation, and medication assessment of numerous medical conditions. Osteoarthritis (OA) is a degenerative joint condition brought on by changes to the bones and cartilage loss in the joints. The examination of bone texture to predict early OA of the knee is a challenging task in processing of medical image. It’s better for viewing severity conditions in medical field rather than to listening. Therefore, it is essential for early prediction and treatment of Knee OA by examining the X-ray images which are medically categorized by experienced doctors by Kellgren and Lawrence (KL) scoring method. The present work suggests a strategy for automated osteoarthritis of the knee classification based on Deep Convolutional Neural Networks (DCNN).The X-ray pictures from Osteoarthritis Initiative (OAI) Dataset are utilized for assessment. Then, a strategy is proposed to extract the features using pretrained Efficient Net architectures such as Efficient Net B0, Efficient NetB1, Efficient Net V2B0, Efficient Net V2B1, Efficient Net V2B2 and Efficient Net V2B3. These eradicated characteristics are then decreased in dimension by Principal Component Analysis (PCA). The reduced features are classified with Basic Convolution Neural Network (CNN) architecture which resulting in a testing accuracy of 95.71% and an imbalanced accuracy of 86.41% using the features of Efficient Net V2B1.