Title: Automated KL grading of knee X-ray images using convolutional neural network

Authors: S. Rajkumar; V.A. Sairam; R. Saranya; N. Sandhiya; V. Shivanie; V. Sapthagirivasan

Addresses: Department of Biomedical Engineering, Rajalakshmi Engineering College, Chennai, 602105, India; Centre of Excellence in Medical Imaging, Rajalakshmi Engineering College, Chennai, 602105, India ' Department of Biomedical Engineering, Rajalakshmi Engineering College, Chennai, 602105, India; Centre of Excellence in Medical Imaging, Rajalakshmi Engineering College, Chennai, 602105, India ' Department of Biomedical Engineering, Rajalakshmi Engineering College, Chennai, 602105, India; Centre of Excellence in Medical Imaging, Rajalakshmi Engineering College, Chennai, 602105, India ' Department of Biomedical Engineering, Rajalakshmi Engineering College, Chennai, 602105, India; Centre of Excellence in Medical Imaging, Rajalakshmi Engineering College, Chennai, 602105, India ' Department of Biomedical Engineering, Rajalakshmi Engineering College, Chennai, 602105, India; Centre of Excellence in Medical Imaging, Rajalakshmi Engineering College, Chennai, 602105, India ' Department of Biomedical Engineering, Rajalakshmi Engineering College, Chennai, 602105, India; Medical Devices and Healthcare Technologies Department, IT Service Company, Bengaluru, 560066, India

Abstract: Knee osteoarthritis affects people across the globe; Kellegren Lawrence's grading method is widely used for diagnosing and grading the diseased condition based on X-ray images. The work aims to develop an AI tool trained by deep learning (DL) algorithms to perform automated classification of the grades of deterioration using knee X-ray image. A modified version of Inception-ResNet-v2, which uses transfer learning, is developed as a CNN model to classify the KL grade of knee X-ray images. Open source OAI dataset containing X-ray images used 9,786 images with ground truth labelling. A web-based AI tool is developed to categorise knee X-ray images into one of five KL grades. The classifier developed and trained on the OAI dataset (curated) produced 75% validation accuracy, 0.74 validation loss, 78% specificity, 69% sensitivity, and 0.942 AUC. The proposed model is helpful to clinical professionals to know the knee osteoarthritis condition and improves diagnostic quality.

Keywords: KL grading; convolutional neural network; CNN; transfer learning; knee X-ray; knee osteoarthritis.

DOI: 10.1504/IJBET.2024.138970

International Journal of Biomedical Engineering and Technology, 2024 Vol.45 No.3, pp.198 - 215

Received: 30 May 2023
Accepted: 27 Sep 2023
Published online: 05 Jun 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article