Feature Extraction of EMG Interference Pattern based on Wavelet Features

E. Abel, H. Meng, and A. Forster (UK)

Keywords

EMG, Wavelet Transform, Feature Extraction

Abstract

Clinical electromyography (EMG) interference pattern (IP) signals are believed to carry more diagnostic information than motor unit action potential (MUAP) alone. Significant features have been obtained from EMG IP signals by a wavelet transform method using the Inter Scale Wavelet Maximum (ISWM) parameter to locate and characterize the rising edges of MUAPs in the IP signal from their propagation at many frequency bands and a parameter representing the singularity of the signal. These wavelet features showed highly significant differences between healthy, myopathic and neuropathic subjects. In this paper, four feature extraction methods were used for further feature extraction from these wavelet features. They are Principle Component Analysis (PCA), Linear Discriminant Analysis (DA), Kernel PCA and Kernel Discriminant Analysis (KDA). The new features gave highly significant differences between healthy, myopathic and neuropathic subjects.

Important Links:

Go Back