Vibroarthrographic Signal Classification for Knee Joint Disorder Detection using Tunable Q-factor Wavelet Transform based on Entropy Measures

Abstract

Listen

Translate article

Research on Vibroarthrographic (VAG) signals presents a promising means for the early diagnosis of knee joint disorders. However, the classification problem for these signals faces serious issues due to their complex and dynamic nature. This study proposes a novel method for decomposing and analyzing VAG signals based on a Tunable Q-factor Wavelet Transform (TQWT) and entropy-based measures. TQWT is used to preprocess and decompose VAG signals recorded during knee motion into subbands. Different entropy metrics, such as approximate entropy, sample entropy, fuzzy entropy, slope entropy, and so on, were computed over different subbands of the signal to capture significant signal features. Effective features were selected using Recursive Feature Elimination (RFE) and then classified using ensemble classifiers such as XGBoost, Ensemble Random Forest (ERF), and RF-logistic regression. The classification accuracy of the proposed sample entropy method was 87.64% and had 90% sensitivity, 86.36% specificity, and 0.88 AUC-ROC. These results demonstrate the ability of the TQWT-based approach to discriminate knee joint abnormalities. Future work will explore performance scaling with larger datasets and apply it to other joint disorders.