Analysis of motor performance in Parkinson's disease through LTI dynamical systems

• Main Author: Ashoori, Ahmad
• Language: English
• Published: University of British Columbia 2014
• Online Access: http://hdl.handle.net/2429/51518

Motor performance in Parkinson's disease is marked by a variety of impairments, and is indicative of faulty feedback mechanisms in the brain that arise due to a lack of dopamine. Quantifying motor performance as well as investigating the correlation of motor behaviour and clinical measures has been sought as in important avenue to study disease pathophysiology. We propose the use of linear dynamical systems to finely characterize motor performance, and for possible interpretations of neurological phenomena, such as compensatory mechanisms that can be cast in the framework of control theory. While coarse measurements of motor performance exist (e.g., maximum velocity, mean velocity, root mean square error), they often cannot distinguish between qualitatively different types of movement. (Consider for example, the wide variety of step response behaviors possible in second order systems which may have the same settling time, but a wide range of damping ratios.) In contrast, LTI models provide a dynamical mapping of the sensorimotor transformation required in tracking tasks. Model parameters of linear dynamical systems (such as damping ratio, and natural frequency) can more readily describe qualitatively similar motor phenomenon than existing coarse measurements. From time-series data of various manual tracking experiments, we first perform a set of preprocessing techniques dependent on the particular experimental setup. The pre-processing techniques have been carefully evaluated to appropriately address possible noise, nonlinearities, and other data irregularities. We then apply existing tools for system identification (e.g., ARX, ARMAX, subspace methods) to identify numerically robust second-order LTI system models. The choice of system identification method (as well as choice of pre-processing techniques) is critical in determining the variability in the resulting model parameters. We identify parameters of the LTI models which are amenable to comparison across subjects (and across groups), and evaluate their statistical significance. The main contributions of this thesis are: - Selection of pre-processing and system identification techniques for three experimental setups for manual tracking experiments - Control theoretic framework for tremor as a compensatory mechanism that is advantageous in some tracking tasks - Sensor-based assessment of rigidity via decay rate - Assessment of cognitive inflexibility through mode detection and delay === Applied Science, Faculty of === Electrical and Computer Engineering, Department of === Graduate