Exploration of Coarse-Graining and Threshold Selection of Lempel–Ziv Complexity on Vibroarthrographic Signals

Lempel–Ziv complexity (LZC) is extensively used for measuring the non-stationary nature of bio-signals like Electrocardiogram (ECG), Electroencephalogram (EEG), Electromyogram (EMG), etc. But LZC is not yet used for analyzing Vibroarthrography (VAG) signals. We study the performance of LZC as compared to other non-stationary measures used in literature like fractal dimension and approximate entropy and observed that LZC offers superior performance. The separability among the classes offered by LZC depends on the selection of threshold used in coarse-graining technique. There are no clear guidelines regarding the selection of threshold for coarse-graining in VAG signals. We compare the amplitude-based and gradient-based coarse-graining techniques and mean and median thresholds in amplitude-based coarse-graining in terms of the ability of LZC measure to distinguish normal and abnormal VAG signals. We also study the effect of the selection of threshold in amplitude-based coarse-graining. We observed that mean-based coarse-graining techniques offer Bhattacharya distance = 0.1609, relative entropy = 1.6835, and p-value = .0027 using Wilcoxon rank-sum test between the normal and abnormal VAG signals. Mean-based coarse-graining techniques provide slightly higher class separability as compared to median-based coarse-graining. The distance measures and p-values indicate that gradient-based coarse-graining is not suitable for the screening of VAG signals.