An affordable and easy-to-use tool to diagnose knee arthritis using knee sound

Arthritis, the most common form of osteoarthritis, affects millions of individuals worldwide. While it can affect any joint in the body, it predominantly affects the hands, neck, back, knees, and hips. Traditional imaging technologies like X-rays are impractical for diagnosing knee pathology and are insufficient for detecting disease symptoms, especially in the early stages of damage. Invasive procedures such as arthroscopy are often required to identify changes, but they are unsuitable for repeated assessments, follow-ups, or monitoring. Vibro-acoustography (VAG), which utilizes vibration and acoustic signals from the human knee during movement, has emerged as a non-invasive diagnostic tool. However, muscle-induced low-frequency irregularities triggered by knee movement significantly impact the accuracy of VAG signals. In this study, a novel machine learning -based approach is introduced for detecting knee osteoarthritis by processing sound signals recorded by a condenser microphone. Knee sounds from both healthy and osteoarthritic patients were captured during knee swing motions (flexion and extension). Three features were extracted from the knee sounds and classified using the Perceptron, MLP, SVM, GRNN, and RBF classifiers. The performance of different features and classifiers was compared, revealing that one feature demonstrated robust results with a high accuracy of 87% and an F1-score of 0.86. Compared to other conventional methods, the proposed approach is characterized by its speed, afford-ability, portability, and user-friendliness, with the potential for easy development into a mobile application.

Automated summary

Arthritis, a common condition affecting millions of people worldwide, can be challenging to diagnose accurately. This study introduces a novel machine learning-based approach using sound signals to detect knee osteoarthritis, offering high accuracy, speed, and affordability compared to traditional methods.