Double Density Dual-Tree Complex Wavelet Transform-Based Features for Automated Screening of Knee-Joint Vibroarthrographic Signals

Pathological conditions of knee-joints change the attributes of vibroarthrographic (VAG) signals. Abnormalities associated with knee-joints have been found to affect VAG signals. The VAG signals are the acoustic/mechanical signals captured during flexion or extension positions. The VAG feature-based methods enable a noninvasive diagnosis of abnormalities associated with knee-joint. The VAG feature-based techniques are advantageous over presently utilized arthroscopy which cannot be applied to subjects with highly deteriorated knees due to osteoarthritis, instability in ligaments, meniscectomy, or patellectomy. VAG signals are multicomponent nonstationary transient signals. They can be analyzed efficiently using time–frequency methods including wavelet transforms. In this study, we propose a computer-aided diagnosis system for classification of normal and abnormal VAG signals. We have employed double density dual-tree complex wavelet transform (DDDTCWT) for sub-band decomposition of VAG signals. The \(L_2\) norms and log energy entropy (LEE) of decomposed sub-bands have been computed which are used as the discriminating features for classifying normal and abnormal VAG signals. We have used fuzzy Sugeno classifier (FSC), least square support vector machine (LS-SVM), and sequential minimal optimization support vector machine (SMO-SVM) classifiers for the classification with tenfold cross-validation scheme. This experiment resulted in classification accuracy of 85.39%, sensitivity of 88.23%, and a specificity of 81.57%. The automated system can be used in a practical setup in the monitoring of deterioration/progress and functioning of the knee-joints. It will also help in reducing requirement of surgery for diagnosis purposes.