A multi-rate and auto-adjustable wavelet decomposition framework for pathological hand tremor extraction

Abstract

Estimation/extraction of pathological tremor signal is of paramount importance in different rehabilitation and clinical applications as it assists the physicians to evaluate the treatment response and investigate undiscovered aspects of tremor disorder. This paper is particularly motivated by the crucial need for having real-time and accurate estimate of the tremor signal in robotic rehabilitation systems to filter out (attenuate) the tremor from the voluntary movement of the patient. We propose a novel, multi-rate and auto-adjustable wavelet decomposition framework, referred to as the MAWD, for pathological hand tremor extraction. The proposed framework consists of two schemes running in parallel, i.e., hyper-parameter adjustment (HPA) scheme and real-time tremor prediction (RTP) scheme. More specifically the aforementioned schemes adaptively decompose the measurement signal consisting of both tremor and voluntary motion into different levels of approximation and use polynomial extrapolation to predict the value of the desired signal for the next time instant. The performance of the proposed framework is evaluated based on real pathological tremor data. The results indicate that the MAWD provides improved accuracy in comparison to the recently developed state-of-the-art pathological tremor estimation methodology.