Thesis

Data science enabled rehabilitation

Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2020
Thesis identifier
  • T15869
Person Identifier (Local)
  • 201772434
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • Stroke is a main cause of impairment/disability. More stroke survivors undergo unsupervised home rehabilitation. Autonomous self-rehabilitation systems using sensing and machine learning are not tailored to patients’ needs. Based on a systematic narrative literature review, home-based rehabilitation systems were taxonomized and new design criteria were formulated for increased patient engagement enhancement and individualism. No system that addresses all the criteria was found in literature. An in-house low-cost home-based rehabilitation Ambient Intelligence (AmI) system was deployed meeting the criteria, and an accuracy evaluation method proposed, in line with medically approved tests. The Timed Up and Go (TUG) and Five Time Sit To Stand (FTSTS) tests evaluate daily living activity performance in the presence/development of comorbidities. The AmI-driven system complies with Accountability, Responsibility, and Transparency (ART) requirements for wider acceptability. A method is presented for generating synthetic datasets complementing experimental observations mitigating bias present due to practical limitations. Also, an incremental hybrid machine learning algorithm is proposed. It combines ensemble learning and hybrid stacking using extreme gradient boosted decision trees and k-nearest neighbours to meet individualisation, and ART requirements while maintaining low computation footprint. The proposed approach was based on the criteria: nonintrusive, nonwearable, motivation and engagement enhancing, individualized, supporting daily activities, cost-effective, simple, and transferable. The motivation method, suitability for elderly, and intended use were examined as supplementary criteria. Indicators of enhanced motivation and engagement, through questionnaire responses, demonstrate that >83% of participants support the proposed system’s motivation and engagement enhancement. The system is fit for purpose with statistically significant (ϱc>0.99, R2 >0.94, ICC>0.96) and unbiased correlation to the gold standard. The model reaches up to 100% accuracy for FTSTS and TUG in predicting associated patient medical condition, and 100% or 83.13%, respectively, in predicting area of difficulty in the segments of the test. Results show an improvement of 5% and 15% for FTSTS and TUG, over previous intrusive approaches. Keywords: Home-based rehabilitation systems, Stroke rehabilitation, Telerehabilitation, Patient participation, Motivation, Comparative effectiveness research, Automated timed up and go test, Automated five time sit to stand test, Self-evaluation, Evaluation of sensor systems, Non-intrusive sensing, Sensing for health, Accountable Artificial Intelligence, Responsible Artificial Intelligence, Transparent Artificial Intelligence, Hybrid ensemble learning, Patient-centric individualised rehabilitation
Advisor / supervisor
  • Stanković, Vladimir
Resource Type
Note
  • Previously held under moratorium from 10th June 2021 until 12th June 2023.
DOI
Date Created
  • 2021
Former identifier
  • 9912984092102996

Relazioni

Articoli