Please use this identifier to cite or link to this item:
Title: Novel features in accelerometer-based gait analysis for long-term monitoring of Parkinson’s disease : a signature of gait.
Authors: Dunne-Willows, Michael James
Issue Date: 2022
Publisher: Newcastle University
Abstract: Parkinson’s Disease (PD) is a neurodegenerative disease that can lead to restricted or slowed movement, gait impairments and increased risk of falling. Over recent decades, instrumented gait analysis (IGA) has contributed much to the understanding of gait impairments in PD. Due to the complexity of gait and high clinical interest a plethora of features have been suggested for gait analysis in the literature pertaining to several groups such as: traditional spatio-temporal (e.g. gait speed), frequency domain, etc. A subset of these traditional gait features has been proposed and validated in PD and older adults as a comprehensive model of gait comprising five factors: pace, rhythm, asymmetry, variability, and postural control. Analysis of gait may be grouped into the assessment of two types of variability, namely, within-subject variability which is needed for personal disease management and inter-subject variability which is useful in quantifying the overall impact of PD on gait. Advances in wearable technology have led to much smaller devices (e.g. accelerometers) being commercially available in conjunction with greatly increased battery lives to the degree that not only lab-based but also continuous recordings over 7 days (real-world) are possible. Wearable technology-based gait analysis is indeed emerging as a powerful tool to detect early disease and monitor progression. Data recorded as part of the ICICLE-GAIT 1 study provides acceleration data for over 100 people with PD and age-matched control subjects in both lab and realworld conditions. These datasets form the basis for the development of a new Phase plot methodology for gait analysis in PD. In this thesis I present a novel methodology for both assessing PD and tracking individual disease progression over multiple timescales. To accomplish this, I introduce a new feature domain, the Phase domain, based on a particular type of recurrence plot known as a Poincar´e plot. Poincar´e plots are sometimes referred to in the literature as return maps, self-similarity plots or Phase plots. Phase plots were being used in the early 1990s in ECG studies to produce self-similarity plots of beat-to-beat intervals. This technique proved to be reliable in detecting atrial fibrillation. The rare instances of its application to other fields are very limited and do not demonstrate any modification or development beyond that which has been used in ECG studies for decades. I develop methodology for application to gait analysis and, indeed, any cyclical biosignals. In this thesis I used the data from the ICICLE-GAIT study to demonstrate that with specific modifications and newly identified features (comprising the Phase domain), this novel Phase plot methodology is highly applicable to gait analysis within PD and provides a framework for: (i) identifying and characterising PD and (ii) individual disease tracking over the years following diagnosis. Throughout these analyses, traditional gait features serve as an established reference and benchmark. I employ statistical methods, such as non-linear mixed effects models and Statistical Parametric Mapping, to model PD progression and assess the clinical utility of Phase plots. I also used Discrete-Time Markov chain modelling, longitudinal analyses, and functional principal components analysis to demonstrate that Phase plots provide an objective, personalised, and clinically relevant signature of gait. In the case of PD patients (and controls to a lesser extent) four distinct Phase plot Types emerge and occur with high within-subject reproducibility, hence the signature interpretation. Many features within the Phase domain proved to be highly sensitive to the disease (people with PD versus controls). Using lab-based data, the Phase domain features outperformed traditional spatio-temporal features in classifying PD. Each domain of features performed similarly well in the prediction of MDS-UPDRS 2 (a useful proxy for PD progression). Specifically, part III of the UPDRS scale was used as this relates to motor function. In real-world conditions Phase plot features showed sensitivity to disease state and physical capability across multiple timescales e.g., daily fluctuations, and also across 18-month follow up time points. The Phase plot-based signature of gait is validated under lab-based conditions to reflect participants’ capacity for gait as well as under real-world conditions as a compact means of monitoring PD and walking performance through gait.
Description: PhD Thesis
Appears in Collections:School of Mathematics, Statistics and Physics

Files in This Item:
File Description SizeFormat 
Dunne-Willows M J 2022.pdf15.72 MBAdobe PDFView/Open
dspacelicence.pdf43.82 kBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.