Publication Date
2025
Document Type
Thesis
Committee Members
Fathi Amsaad, Ph.D. (Advisor); Wen Zhang, Ph.D. (Committee Member); Huaining Cheng, Ph.D. (Committee Member)
Degree Name
Master of Science (MS)
Abstract
Bilateral idiopathic carpal tunnel syndrome (CTS) is a neuromuscular disorder characterized by compression of the median nerve at both wrists, leading to symptoms such as pain, numbness, tingling, and muscle weakness. Unlike unilateral cases, bilateral idiopathic CTS presents distinct therapeutic challenges due to the simultaneous involvement of both hands and the lack of an identifiable underlying cause. This study explores the application of machine learning techniques to predict the optimal sequence of physiotherapeutic interventions Stretching followed by Myofascial Mobilization (S/M) or the reverse (M/S) in female patients with bilateral idiopathic CTS and right hand dominance. Data were drawn from a randomized controlled crossover trial involving 73 participants, with male and left-handed individuals excluded to maintain a homogenous sample. The dataset underwent a comprehensive preprocessing pipeline, including feature engineering through Principal Component Analysis (PCA) of strength measurements. Several machine learning models were evaluated based on predictive performance, focusing on features such as PCA-derived strength components, electrophysiological severity scores (e.g., Bland and Padua classifications), and sensory assessment results. CatBoost emerged as the top-performing model, achieving a ROC-AUC of 0.985 and a test accuracy of 96.5%, and was selected for final implementation. To assess the security and robustness of the ML-driven system, adversarial attacks were simulated by introducing targeted perturbations within the PCA feature space. Despite these perturbations, the CatBoost model maintained an adversarial accuracy of 94.01% with an attack success rate of only 5.99%, demonstrating resilience against subtle, clinically plausible input changes. Further experimental evaluation indicated that both CatBoost and Random Forest models sustained high classification accuracy under adversarial conditions. The Random Forest model achieved an adversarial accuracy of 96.83%, with a limited attack success rate of 3.17%, suggesting that low-magnitude perturbations had minimal impact on model performance. Analysis of feature importance highlighted the critical role of the Strength_PCA1 and Strength_PCA2 components in model predictions. Additionally, explainability techniques confirmed that the models consistently relied on clinically meaningful patterns, reinforcing their potential utility in medical decision making. Overall, the findings of this research demonstrate the strong feasibility of AI driven approaches for individualized treatment planning in CTS rehabilitation. Moreover, the resilience of the developed ML models against latent-space adversarial attacks underscores their robustness, security, and practicality for deployment in clinical decision support systems.
Page Count
89
Department or Program
Department of Computer Science and Engineering
Year Degree Awarded
2025
Copyright
Copyright 2025, all rights reserved. My ETD will be available under the "Fair Use" terms of copyright law.
