A Proximity–Margin Hybrid Learning Architecture for IoT-Enabled Vibroarthrographic Signal Analysis in Knee Health Diagnostics

Abstract

Knee osteoarthritis (OA) and ligament injuries are among the most common musculoskeletal disorders affecting mobility and quality of life, particularly in aging populations. Early diagnosis and continuous monitoring are critical for effective treatment, but traditional diagnostic methods, such as X-rays and MRI, often detect abnormalities only at advanced stages, making early intervention challenging. These conventional techniques are also expensive, time-consuming, and require specialized facilities and expertise, which limits accessibility for regular monitoring. Vibroarthrography (VAG) has emerged as a non-invasive approach that records acoustic and vibrational signals generated by the knee joint during movement, providing valuable insights into joint health. However, interpreting VAG signals manually is difficult and prone to errors, and standalone clinical evaluation cannot capture the complexity of signal patterns effectively. This research addresses these limitations by integrating VAG with machine learning algorithms to classify knee conditions into three target classes: normal, osteoarthritis, and ligament injury. The system employs models such as Support Vector Machine classifier (SVM), KNearest Neighbours Classifier (KNN), Decision Tree Classifier (DTC), and a hybrid stacking model combining SVM and KNN also known as KneeSense FusionNet Model. The proposed system provides a user-friendly interface for clinicians and patients, enabling real-time, accurate assessment and continuous monitoring of knee health. Its significance lies in offering a cost-effective, non-invasive, and reliable diagnostic tool that facilitates early detection, personalized treatment, and improved patient outcomes, while overcoming the limitations of traditional diagnostic methods.