A Kinematic Framework for Upper Extremity Rehabilitation Assessment : Expectation- Maximization as a Motor Learning Model

Résumé

Motor learning as a recovery mechanism is assumed to be a framework that drived and guided physical therapy and now since the advent of robotics doing the same to the rehabilitation devices. The rehabilitation process presents the intersection of many different interconnected facets that co-interact to produce recovered movements. The use of the technology introduces many benefits while contributing to the complexity of the phenomena at hand. We interest our research to the passive exosquelette training of the upper limb. We propose an adaptive intra patient assessment scale that is capable of detecting intra-patient performance changes during robotic training. Motor learning, the process of our brain's acquiring newer motor skills or relearning those he lost due to neurological or traumatic incident is our portal to investigating this phenomenon. The interaction of the system that is composed of the device, the incentive in form of exercise games and the patients with all its level of existence, physiological, psycho-logical, and cognitive is the system of study. The components present heterogeneous qualities and dynamically driven changes. The system output in the form of the trajectories executed is our gauging instrument to investigate the interactions within the system. We formulate the trajectory model as a Markov Chain and use the Kalman Filter to estimate the smoothed states. While dynamics are variant in time we model the assumptions about the movement into a dynamical formulation and estimate its parameters from data. To account for the time variability we introduce parallel noise source to the dynamics and estimate it using an Expectation-Maximization algorithm. The temporal nature being only a single facet of the kinematic phenomena, we assume a variable temporal alignment and estimate it using Expectation-Maximization iteration to increase the likelyhood of the estimated model compared to the observed trajectories. Once learned the model dependent and extracted parameters are used to compare between differences in performance. The properties of the clinical assessment tools are investigated and results are formulated to answer the commonly reported needs. Stemming from the same fundamentals of motor learning, we aimed to define a new visual assessment instrument that is intended to fulfill the need of patient-first easily communicated feedback form. We present and assess clinical properties of the tools while providing validating results on clinical data attesting the longitudinal sensitivity of the tool. The underlying assumption of the visualization was then assessed using an objective measure of maximum probability value derived using a probabilistic model of the trajectories and expected on a highly likely trajectory model learned using a Kernel-Near-Neighbors Regressor.

Citer ce document

Accès au document

Ce document est hébergé sur son dépôt institutionnel d'origine.

Auteur(s)

Statistiques

Consultations : 2

Téléchargements : 0