A quantitative classification of essential and Parkinson's tremor using wavelet transform and artificial neural network on sEMG and accelerometer signals

Correct discrimination of essential tremor from Parkinson's tremor is a major problem in clinical neurology as minor differences in the tremor patterns are hard to distinguish. Mathematical analysis of tremor signals recorded non-invasively has been widely accepted for tremor differentiation. However, classification of tremor signals collected from electromyograph or accelerometer, based on time and frequency domain techniques has limited accuracy because of overlapping frequency range and non-stationary nature of those signals. This paper describes a simple, non-invasive decision making logic method for discrimination of tremor. Wavelet transform based feature extraction technique in combination with feed forward type artificial neural network is proposed. Fractal dimensions of wavelet features of the decomposed detailed coefficients are used as the feature matrix. The neural network classified the tremor sEMG signals with 91.66% accuracy and 100% in case of accelerometer signals. Although, the classification accuracy of sEMG signal is comparable to that of accelerometer but the localized involuntary vibratory nature of tremor at the extremities of human body puts accelerometer as a better option in cases where tremor fails to excite the muscle. This proposed classification algorithm adds strength to the non-invasive signal detection methods at reduced cost and higher sensitivity.