Detection of manual tracking submovements in Parkinson's disease through hybrid optimization

Abstract

Seemingly smooth motions in manual tracking, (e.g., following a moving target with a joystick input) are actually sequences of submovements: short, open-loop motions that have been previously learned. In Parkinson's disease, a neurodegenerative movement disorder, characterizations of motor performance can yield insight into underlying neurological mechanisms and therefore into potential treatment strategies. We focus on characterizing submovements through Hybrid System Identification, in which the dynamics of each submovement, the mode sequence and timing, and switching mechanisms are all unknown. We describe an initialization that provides a mode sequence and estimate of the dynamics of submovements, then apply hybrid optimization techniques based on embedding to solve a constrained nonlinear program. We also use the existing geometric approach for hybrid system identification to analyze our model and explain the deficits and advantages of each. These methods are applied to data gathered from subjects with Parkinson's disease (on and off L-dopa medication) and from age-matched control subjects, and the results compared across groups demonstrating robust differences. Our preliminary results suggest that the three-mode switched system could be extended with parameterization of the dynamics subject to a stochastic reset map.