Use this URL to cite or link to this record in EThOS:
Title: Enabling real-time automatic assessment of patient exercises for technology-assisted physical rehabilitation interventions
Author: Sarsfield, J.
ISNI:       0000 0004 8502 4713
Awarding Body: Nottingham Trent University
Current Institution: Nottingham Trent University
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Access from Institution:
Technology-assisted physical rehabilitation interventions (TAPRI) have the potential to offer patients a safe, motivating and always accessible platform for undergoing rehabilitation. The emergence of compact and affordable depth sensors provide an opportunity to realise such interventions in a home environment. These types of depth sensors can run pose estimation algorithms that track full-body human joint positions in real-time. TAPRIs that provide real-time patient performance assessment and feedback require sufficiently accurate algorithms to ensure a correct assessment. The research presented in this thesis aims to overcome some of the algorithmic challenges in enabling real-time patient performance assessment and feedback. This research focuses on two algorithms: real-time tracking of human joint positions and real-time segmentation of exercise repetitions. This research targets stroke rehabilitation as a challenging use case for achieving real-time patient exercise assessment as stroke patients often have varying levels of mobility. Research contributions. The first contribution of this thesis is a quantitative and clinical evaluation of a state-of-the-art pose estimation algorithm (human joint tracking) to determine if the joint position estimations are sufficiently accurate for correctly assessing stroke rehabilitation exercises. This evaluation also determines what the limitations are and propose recommendations for future pose estimation algorithms intended for clinical applications. The second contribution is an evaluation of the inter-rater reliability of clinicians assessing the suitability of the pose estimation algorithm, to quantitatively determine where the clinicians are in agreement and propose more robust criteria for the assessment of new clinical technologies. The final contribution is the proposal of a real-time segmentation algorithm that requires only a single exemplar repetition of an exercise to segment repetitions from other subjects, including those with impaired mobility. Main research findings and results. The accuracy of current state-of-the-art pose estimation algorithms are insufficient for correctly assessing patient performance. There was a low inter-rater agreement between clinicians evaluating the accuracy of the individual joints of a state-of-the-art pose estimation algorithm, however overall the accuracy was found to be insufficient. Our proposed segmentation algorithm correctly segments 90% of stroke patient exercise repetitions from our own rehabilitation exercise dataset and is capable of segmenting a 20 second window at 30Hz in real-time on a desktop computer.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available