Tremor quantification through data-driven nonlinear system modeling

Tremor is a repetitive uncontrollable movement of a body part that represents a cardinal symptom in several widespread neurological diseases. Being often related to a lifelong incurable condition, tremor has to be monitored in order to follow disease status and adjust the therapy. This paper proposes a novel method of tremor quantification by modeling the repetitive movement underlying tremor as a periodic solution to an autonomous nonlinear system whose parameters are estimated from the data. A 3D fused acceleration measurement in a smart phone is utilized for data acquisition. The measured trajectory is seen as a superposition of a slow voluntary translational-rotational movement and a involuntary repetitive component produced by autonomous nonlinear dynamics. The repetitive part of the data is extracted by an event-based procedure relying on filtering and scaling of a position estimate obtained from the measurements. A planar representation of the underlying dynamics is used for estimating the parameters of a second-order system with a polynomial nonlinearity. Then, the tremor amplitude is captured by the scaling coefficients while the frequency content is defined by the nonlinear dynamical model. The efficacy of the proposed quantification approach is demonstrated on experimental data.