Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

Data-driven Rehabilitation Development for Stroke Patients Using Machine Learning Techniques

Allen, Marcus (2020) Data-driven Rehabilitation Development for Stroke Patients Using Machine Learning Techniques. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

Download (10MB) | Preview


A portable motion capture system that can guide a stroke patient’s rehabilitation session will help them regain motion in their arm quicker and save money on medical bills. The goal of this dissertation was to investigate stroke patient motions to lay the groundwork for such a system. Specifically, the dissertation focused on developing an automated system for stroke diagnostics and task-oriented rehabilitation. This form of telerehabilitation is new to the field since previous systems do not create skill-based instructions for the patient.
The system created in this dissertation was demonstrated to classify a patient’s stroke severity using supervised machine learning techniques. It was shown how this information can be used to create a personalized rehabilitation protocol for that patient. For this study, inertial measurement units were used to collect joint kinematics and kinetics during reaching tasks for 10 healthy and 17 stroke subjects, and from that data machine learning features were defined for a support vector machine classification algorithm. These parameters were validated against a gold standard optical tracking motion capture system (Optitrack, NaturalPoint, Inc.) where the joint results were calculated using the Motion Monitor Biomechanics engine. The kinematic and kinetic features extracted from these patients for the machine learning model also served a dual purpose by creating a hierarchal rehabilitation session by categorizing reaching tasks by difficulty. The features used to determine a patient’s task hierarchy for rehabilitation were determined based on the support vector machine algorithm that classified stroke severity. Task rankings were developed for stroke individual stroke
subjects as well as for classes of stroke subjects. The rankings were unique for classes and for individuals, but their general orders coincided with intuitive task difficulty.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Allen, Marcusmca36@pitt.edumca36
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairClark, Williamwclark@pitt.eduwclark
Committee MemberCole, Danieldgcole@pitt.edudgcole
Committee MemberSethi, Amitasethi@pitt.eduasethi
Committee MemberSharma,
Date: 3 August 2020
Date Type: Publication
Defense Date: 8 November 2019
Approval Date: 3 August 2020
Submission Date: 16 April 2020
Access Restriction: 1 year -- Restrict access to University of Pittsburgh for a period of 1 year.
Number of Pages: 242
Institution: University of Pittsburgh
Schools and Programs: Swanson School of Engineering > Mechanical Engineering
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: motion capture, occupational therapy, mechanical engineering
Date Deposited: 03 Aug 2021 05:00
Last Modified: 03 Aug 2021 05:15


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item