Development of An Artificial Neural Network (ANN) Model for Gait-Based Classification of Neurodegenerative Disorders (NeuroGaitNet)
Open Access
- Author:
- Mohamed, Mohamed
- Graduate Program:
- Mechanical Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- October 03, 2024
- Committee Members:
- Rick Ciocci, Thesis Advisor/Co-Advisor
Aldo W Morales, Committee Member
Brian Maicke, Professor in Charge/Director of Graduate Studies
Anilchandra Attaluri, Committee Member - Keywords:
- Keywords: Neural Network
Gait Analysis
Neurodegenerative Disorders
Power Spectral Density (PSD)
Data Augmentation
Deep Learning
Huntington's Disease
Clinical Diagnostics
Artificial Intelligence. - Abstract:
- Early diagnosis of neurodegenerative diseases like Huntington's disease (HD), Parkinson's disease, and amyotrophic lateral sclerosis (ALS) is a tough challenge primarily because motor symptoms overlap and are often symptomatic of more than one condition; also, traditional diagnostic methods are costly and demanding. There has been relatively little exploration of gait analysis in Huntington's disease compared to the well-established use in Parkinson's and ALS. This paper aims to fill this gap by proposing a neural network model named NeuroGaitNet based on deep learning approaches, which are used for the classification of gait disorders using gait data collected from wearable sensors. The severity of gait alterations increases with neurodegeneration, but nuanced differences demand sophisticated analysis. NeuroGaitNet can analyze gait's time domain and frequency data and generate stride intervals, footfall contact times, and power spectral density (PSD). The model was trained on ALS, Parkinson's, Huntington's data [14], and healthy controls, and it had an average test and validation accuracy of 95.5% and validation accuracy of 91.97%. Few cases were included, so dataset imbalance was considered, and data augmentation techniques (Gaussian noise injection) were used. This work bridges a critical research gap by presenting a scalable and automatic tool for early classification of neurodegenerative diseases targeting Huntington's Disease (HD). Moreover, NeuroGaitNet has a user-friendly graphical interface for processing gait data in real-time. It provides instant classifications of gait disorders that are functional in research and clinical settings by integrating deep learning with sophisticated gait analytics. This study pushes the boundaries in diagnostic technology for neurodegeneration, including Huntington's disease, bringing a real opportunity to transform pre-symptomatic detection and clinical outcomes.