Login (DCU Staff Only)
Login (DCU Staff Only)

DORAS | DCU Research Repository

Explore open access research and scholarly works from DCU

Advanced Search

Exploring machine learning, real-time bio-feedback, and inertial sensor accuracy for the prevention of running-related injuries

Lillis, Clare (2022) Exploring machine learning, real-time bio-feedback, and inertial sensor accuracy for the prevention of running-related injuries. PhD thesis, Dublin City University.

Recreational running is popular, however, incident rates of running related injuries (RRIs) are very high. Predisposition to injury can be assessed through expensive, laboratory-based biomechanical screening. Wearable wireless inertial sensors offer a potential solution, but accurate orientation data are required. This thesis examined the prevention of RRIs, by aiming to improve sensor accuracy, and investigate applications of biofeedback and machine learning. This thesis explored improving (magnetometer-free) orientation accuracy during running, through examination of (i) Z-axis de-drifting, (ii) data-loss (iii) and modifications to the Madgwick filter. Despite some accuracy improvements (i, iii), overall errors were unsuitable for running based applications. Impact loading is associated with RRIs, with thigh angle (quasi-measure of knee-flexion) potentially important in load attenuation. Loading can be altered directly (loading-based biofeedback) or indirectly (technique-based biofeedback), these two types of biofeedback were compared. A mobile phone application was developed providing audio biofeedback to reduce impact accelerations and encourage a ‘softer’ running technique. Both types of feedback reduced loading at the tibia and sacrum, however, tibia loading reduced better with impact accelerations biofeedback, and sacrum loading with thigh angle biofeedback. It would be beneficial to identify runners who may be predisposed to injury. Seven supervised machine learning models were developed to identify runners who may be likely to sustain RRIs, using inertial, kinematic and clinical data collected on 150 prospectively tracked runners. These models resulted in weak predictive accuracy (0.58-0.61 AUC). As we cannot identify runners predisposed to injury, all runners must be recommended for injury prevention interventions. Orientation accuracy was found to be sufficient for relative measures of running technique in the biofeedback app. Future work could investigate biofeedback app use in relation to reduction of RRIs. Additionally, running injury prediction could be examined further with respect to extracting different features (continuous measures) or predicting specific injuries.
Item Type:Thesis (PhD)
Date of Award:November 2022
Supervisor(s):Moran, Kieran and O'Connor, Noel E.
Uncontrolled Keywords:Biomechanics; Sensors; Running, Injury
Subjects:Biological Sciences > Biosensors
Computer Science > Machine learning
Medical Sciences > Sports sciences
DCU Faculties and Centres:DCU Faculties and Schools > Faculty of Science and Health > School of Health and Human Performance
Research Institutes and Centres > INSIGHT Centre for Data Analytics
Funders:Insight SFI Research Centre for Data Analytics, Science Foundation Ireland (SFI) under Grant Number SFI/12/RC/2289_P2, European Regional Development Fund
ID Code:27688
Deposited On:17 Nov 2022 16:14 by Kieran Moran . Last Modified 17 Nov 2022 16:14

Full text available as:

[thumbnail of Clare Lillis_PhD Thesis_e-copy.pdf]
PDF - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Creative Commons: Attribution-Noncommercial-No Derivative Works 4.0


Downloads per month over past year

Archive Staff Only: edit this record