A machine learning approach to objective measurement of tremor severity in Parkinson's disease: clinical and user perspectives on wearable devices

AlMahadin, G.M.M. ORCID: 0000-0001-9244-1054, 2021. A machine learning approach to objective measurement of tremor severity in Parkinson's disease: clinical and user perspectives on wearable devices. PhD, Nottingham Trent University.

[img]
Preview
Text
Ghayth ALMahadin 2022.pdf - Published version

Download (2MB) | Preview

Abstract

Tremor is the most typical common and simply recognised symptom of Parkinson's disease (PD) and presents in 70% - 90% of PD patients. In addition, tremor severity often indicates PD progress and severity and it can be used to evaluate treatment efficiency. Currently, the severity of Parkinson's tremor is scored based on the Movement Disorders Society's Unified Parkinson's Disease Rating Scale (MDS-UPDRS), however, the MDS-UPDRS is subjective and can be a lengthy process. Advances in wearable technologies combined with Machine Learning (ML) techniques have enabled the development of new approaches for the objective assessment of PD motor symptoms. A limited number of commercial systems are available with limited adoption and implementation due to the apparent lack of clinicians' and patients' perspectives. The goal of this research is to develop and validate a comprehensive solution to measure and quantify PD tremor severity objectively that incorporates the analysis of the perspective of the patients and healthcare professionals and provide an appropriate technology based solution. A holistic approach was adopted comprising of qualitative and quantitative methods divided into three stages. Firstly, a qualitative method using semi-structured interviews identified the perspectives of both healthcare professionals and patients linked to current assessment methods and their requirements for wearable devices. The results showed that a well-known assessment process such as MDS-UPDRS was not used routinely in clinics as it is time consuming, subjective, inaccurate and dependent on patients' memories. Participants suggested that objective assessment methods are needed to increase the chance of effective treatment. The participants' perspectives were positive toward using wearable devices. Healthcare professionals stated a need for an economical solution that provides concise information and is easy to use and interpret and should mimic the current scale. Secondly, a novel framework is proposed to enhance the tremor severity classification. The proposed approach is a combination of signal processing and resampling techniques integrated with well-known classifiers. The results show that over-sampling techniques performed better than other resampling techniques. The proposed approach has solved the imbalanced data problem and it has improved tremor severity detection significantly without neglecting minority classes and achieved 95:04% overall accuracy, 96% G-mean, 93% IBA and 99% AUC with Artificial Neural Network based on Multi-Layer Perceptron (ANN-MLP) with Borderline SMOTE. Finally a recommended system was identified to measure tremor severity. The system comprises of recommended tasks, classier, classier hyper-parameters and resampling technique. In this stage, a novel comprehensive method is developed to discriminate tasks' effect on tremor severity detection by developing an efficient and unique metric rule-based algorithm to identify recommended and non recommended tasks to be performed for tremor data collection. This establishes a novel quantitative framework that is based on an exhaustive sequential filtering algorithm that takes into consideration various combinations based on different advanced metrics instead of depending on a single metric. Results showed that ADL tasks that involve direct wrist movements are not suitable for tremor severity assessment. The findings of this research suggest that the recommended system is the SVM classier combined with Borderline SMOTE over-sampling technique and the tasks are sitting, stairs up and down, walking straight, walking while counting, and standing and achieved 98% accuracy, 98% F1-score, 97% IBA, 98% G-mean and 99% AUC. The novel system solutions and the results presented in this thesis demonstrate a significant contribution towards the objective measurement of tremors in Parkinson's disease. New data is also presented for policy-makers and healthcare professionals which provides new perspectives in relation to the objective assessment of PD in current clinical practice.

Item Type: Thesis
Creators: AlMahadin, G.M.M.
Date: September 2021
Divisions: Schools > School of Science and Technology
Schools > School of Architecture, Design and the Built Environment
Record created by: Linda Sullivan
Date Added: 07 Apr 2022 09:24
Last Modified: 07 Apr 2022 09:24
URI: http://irep.ntu.ac.uk/id/eprint/46070

Actions (login required)

Edit View Edit View

Views

Views per month over past year

Downloads

Downloads per month over past year