Modelling Parkinson's Progression using Naturally Spoken Language Features
Open Access
- Author:
- Nagarapu, Murali Nandan
- Graduate Program:
- Computer Science and Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- June 23, 2021
- Committee Members:
- Mahanth Gowda, Thesis Advisor/Co-Advisor
Bhuvan Urgaonkar, Committee Member
Mahmut Taylan Kandemir, Committee Member
Chitaranjan Das, Program Head/Chair - Keywords:
- Parkinson's Disease
Speech Features
Early Prediction
Progression modeling - Abstract:
- Slow progressing Parkinson’s Disease (PD) is a neurodegenerative disorder that affects the nervous system. It has become one of the world’s most important community health diseases, which is growing incrementally and has impacted several nations. With over 1% of the population being diagnosed with PD, it has become crucial to detect and predict the advancements well advance in time. The majority of impact PD has on an individual’s health lies in motor capabilities and dopamine-generating neurons. It is also observed that over the due course of disease progression, patients lose their grip on speech and suffer limb rigidity, Bradykinesia, and gait problems. Current medical advancements have no cure for PD but could improve symptoms with early detection of the PD onset. In addition, the clinical diagnosis takes the patient’s history and standard physical monitoring into account over a period ranging from weeks to months. This study focuses on strategically supporting clinical methods by reducing the time for identifying and modeling the disease progression in patients without having a restrictive setting to perform the diagnosis. Considering that most Parkinson’s patients are impacted by vocal articulation, this thesis performs speech analysis as it is observed to provide reliable insights about the presence of the PD. The goal of this work is to be able to use the speech features to classify an individual as (a) Healthy (H), (b) Person with Parkinson’s (PWP), or (c) Potential Parkinson’s (PP). A comprehensive set of experiments on variously known speech features such as LPCC, MFCC, and hand-crafted features based on voice coherence are done to model this classification task. These proposed experiments allowed for an in-depth understanding of the impact of these features in modeling PD progression in any individual well before the actual diagnosis. Among the various ML techniques, the best results were achieved within ranges of 80% to 90%. The findings in this study also suggest that the modulation of the speech features over time can also bring essential cues to how the disease progresses in individuals of different categories.