An investigation on the use of a thigh-worn accelerometer for the detection of freezing of gait and fall events

Freezing of gait (FoG), a common and debilitating movement disorder, often emerges in the advanced stages of Parkinson's disease. FoG episodes have serious implications for Persons with Parkinson's (PwP), contributing to falls and fall-related injuries and significantly negatively impacting their psychological well-being and quality of life. In recent years, advancements in machine learning and the miniaturisation of wearable inertial sensors have paved the way for the development of FoG detection algorithms capable of detecting the distinct gait patterns associated with FoG. Accurate FoG detection may lead to a better understanding of the condition and to the development of more effective preventative strategies. PwP affected by FoG experience a significantly elevated risk of falls, with FoG being one of the major risk factors for falls in PD. Wearable sensors provide valuable data that can be analysed retrospectively using machine learning techniques. This postprocessing analysis has the potential to reveal gait patterns associated with falls, leading to an improved understanding of fall risk factors and the development of preventative strategies. This thesis presents an in-depth study on the development and validation of algorithms for detecting FoG and falls among individuals with Parkinson's Disease (PD) using thigh-worn accelerometers. This research encompasses a comprehensive analysis of existing algorithms, detailing their architecture, and the reproduction of these algorithms to confirm their performance as reported in the literature. The validation process utilised multiple publicly available datasets, testing the algorithms using realworld fall and FoG events, with performance metrics in line with the findings of the original studies. After implementing the algorithms as originally defined, potential scope for enhancement in the algorithms was identified, and targeted modifications were applied to the algorithms leading to significant improvements. The enhanced FoG algorithm achieved 87.4% sensitivity and 82.2% specificity. The falls algorithm achieved 99.3% sensitivity and 99.7% specificity on simulate fall data and 100% detection of real-world fall. The successful development of these detection algorithms paves the way for their integration into existing assistive devices like cueing systems. This could offer valuable insights into the nature of PD and facilitate telehealth solutions that improve the quality of life for PwP. The algorithms developed in this thesis have been submitted as an invention disclosure form. There are no publications associated with this thesis.

Publisher

University of Galway

Rights

Attribution-NonCommercial-NoDerivatives 4.0 International

cbn

Collections

University of Galway Theses (Research Masters Theses)