Fractal number and spectral skewness: two features for the pattern classification of motor unit action potentials

Two signal features are proposed to help in the digital pattern classification of motor unit action potentials (MUAPs) from human electromyography (EMG). These are the fractal number and the spectral skewness, which showed promising results in the discrimination between polyphasic and normal action potentials. Signals obtained with a concentric needle electrode (Teca CT-50) were amplified and filtered (bandwidth set at 20-10 kHz) before being recorded by an HP 3960 tape recorder (bandwidth 0-5 kHz). The data acquisition and processing were done with an IBM-PC-XT-compatible computer having a 10-bit analog-to-digital conversion board. The preliminary results indicate that the proposed features may be useful additions to those already described in the literature for computer-based MUAP classifiers. The fractal number may be a powerful wavelet measure for classification purposes. Spectral skewness was found to be sensitive to signal complexity, and will be investigated further