DYNAMIC TIME WARPING ANALYSIS IN POST-STROKE REHABILITATION

I have absolutely no idea what this is, so send your doctor after it. It's only 8 pages long with wonderful mathematical equations in it.
http://biomechanica.hu/index.php/biomech/article/view/197/303
Bálint Magyar1, Gábor Stépán1, I-Ming Chen2
1Budapest University of Technology and Economics, Faculty of Mechanical Engineering,
Department of Applied Mechanics
2Nanyang Technical University, School of Mechanical and Aerospace Engineering
magyar@mm.bme.hu
Abstract
This paper presents the application of the Dynamic Time Warping (DTW) algorithm in the analysis
of human functional movements in activities of daily living (ADLs). Dynamic Time Warping
was originally developed for automatic speech recognition, though the method has been adopted
by several fields of biomechanics. As a part of the post-stroke rehabilitation project COSMOSYS,
the aim is to quantify the ADL performances of hemiparetic subjects, hence to be able to track
their progress during physiotherapy.
Introduction
Dynamic Time Warping in automatic speech recognition is used to measure the similarity of
two audio sequences, which may vary in time and speed. The sequences are warped nonlinearly
in the time dimension to determine the „score” of their similarity independent of
certain non-linear variations in pace. The sequences are warped non-lin-early in the time
dimension to determine the “score” of their similarity independent of cer-tain non-linear
variations in pace for various applications of DTW.1–6 The rehabilitation of hemiparesis
after stroke demanded a comparison method that is able to express the correlation of two
data sets. The aim is to evaluate the measured human functional movements also called
Activities of Daily Living (ADLs), i.e. to qualify those with a single scalar. The difference
of the measured and the reference ADL datasets can be used to evaluate the patients’ performance.
The present approach bridges the gap between the objective sensory information available on
normal and pathological human movements on one side and the subjective qualitative evaluation
of these motions by the skilled professional in the form of performance scales on the
other side. While the latter have already been standardized among clinical professionals, the
error of human motion cognition by objective evaluation is still remarkable: evaluation of human
movements may differ due to the imperfectionof human cognitive capabilities, or simply
from training, practice, institution, location and nationality of the clinical professional. The
error can be radically reduced by the proposed co-iterative analytical-statistical method. Feeding
of the DTW metrics into the robot controller makes the robot driven physiotherapy biomedically
determined. In present approach the DTW metrics have been produced from measurements
by 3D motion analyser whereas the robot can measure the same parameters from
their integrated sensors: the motor encoders.