|
Abstract:
|
Experiments assessing upper extremity motor control produce novel kinematic data sets that capture the multiple hand reach movements within each of many subjects. Subjects reach for a specified target in multiple consecutive trials during which the trajectories of arm movements are recorded. We set out to understand inter- and within-subject differences in motor skills and improvements over repeated trials and how they are related to subject-specific information such as injury or disease, or recovery. In this work, we seek to quantify variabilities in the shape and speed of progression along trajectories via multilevel curve registration (MCR). In a Bayesian framework, each trajectory of arm movements within a subject are assumed to be an affine transformation of an individual-specific mean template curve, which is subject to further time-warping. The affine transformations and warping functions are assumed more similar within each subject a priori to account for individual differences. We show that MCR provides more accuracy variability characterization and less biased downstream regression estimates than alternatives that ignore multilevel structure or temporal misalignment.
|
