Nonparametric Time-Domain Tremor Quantification With Smart Phone for Therapy Individualization

This paper deals with a low-complexity time-domain tremor quantification method for individualization of therapies in medical conditions, where tremor is a cardinal symptom. This method produces a tremor severity estimate based on the measurements acquired from a standard sensor platform of a smart phone during a smooth voluntary movement. The estimate is calculated from a data set recorded over tens of seconds but can also be used for unobtrusive tremor monitoring over long time. Besides tremor amplitude, its frequency is also evaluated over time, thus providing the means to distinguish between, e.g., rest and action tremor. No analytical model is assumed for the tremor signal form. The characterization of tremor severity is performed by the steady-state analysis of a Markov chain, whose states correspond to different intervals of tremor amplitude. The utility of the proposed quantification approach is illustrated by clinical data obtained during Deep Brain Stimulation programming sessions. The results are compared with a conventional approach utilizing spectral analysis to demonstrate the benefits of the proposed method.