Tunable Q - wavelet transform based features for automated screening of knee-joint vibroarthrographic signals

Abstract

Pathological conditions in the knee joint can be caused by vibrations emitted in knee joint while bending or extending the leg. To detect this vibrations or to know the inner condition of the joint like softening, coarseness, breakage or it’s state of lubrication of articular cartilage surface, vibroarthrographic (VAG) signal is useful. The VAG signal is used because of fluctuating and nonlinear property which is helpful to extract the condition of knee joint. In present work on VAG signals by using the TQWT decomposition. VAG signals are partitioned into sub-band signals of distinct frequencies. There are different features of VAG Signals, fluctuating in nature like Kraskov Entropy (KE) and Signal Fractal Dimension (SFD). The performance of feature selection (FS) techniques can be validated through using the Least square support Vector machine (LS-SVM). By using LS-SVM classifier we have achieved an accuracy of 86.91%, sensitivity of 88.33%, specificity of 86.66% based on the input of 89 VAG signals.