Detecting Knee Osteoarthritis using X-ray Images: An Image Processing and Deep Learning Approach

• Sudheer Kumar E

  Author

Knee osteoarthritis (OA) is a disease that affects the articular cartilage of the knee joint of humans. This gradually leads to the deterioration of the knee joints. Loss of flexibility, joint stiffness, swelling, pain, and bone spurs are some of the symptoms of knee OA. Medical image processing and deep learning can be employed for the task of classifying medical images and the detection of certain diseases. In this proposed work, we have collected the Xray images from Kaggle, which is the Mendeley Data, VI. Certain preprocessing techniques such as CLAHE, resizing, and data augmentation have been implemented. The dataset contains 5 classes representing the KL-scores from 0 to 4. Transfer learning is used and pre-trained models such as ResNet50, ResNet152, DenseNet121, MobileNetV2, and InceptionV3 have been trained, validated, and tested on our dataset. The efficiency of the proposed approach has been evaluated by carrying out experiments on two sets of multi-class datasets containing 5 and 3 classes. The original dataset containing five classes was transformed into a new dataset containing three classes by merging some of the closely related classes. It was found that InceptionV3 model performed the best on both the sets of data, i.e., the one with five classes and the other one with three classes. The accuracy achieved by InceptionV3 on the 5-class dataset was 67.15%. It gave an accuracy of 76.75% on the dataset containing 3 classes (healthy knee, moderate knee OA, and severe knee OA).

• PDF