Investigation of Machine Learning Techniques to Aid in the Diagnosis of Neurodegenerative Diseases

Abstract

The thesis summarized in this document introduces alternative, rapid, low-cost, and effective solutions, aided by machine learning techniques, to support the diagnosis and differentiation of neurodegenerative diseases (NDDs) such as Parkinson’s Disease, Huntington’s Disease, and Amyotrophic Lateral Sclerosis. These diseases, characterized by the progressive loss of neurons, have no cure, and diagnosis is predominantly clinical. By leveraging novel features extracted from gait signals through dynamic fluctuation analysis and harmonic distortion, the thesis achieves highly accurate results with specificity and sensitivity ranging from 96% to 100% for automatic NDD classification, serving as a diagnostic aid system. Furthermore, it presents and discusses an innovative approach to NDD diagnosis focused on the patient’s well-being, aiming to reduce examination duration and physical effort required for gait signal collection. These contributions represent innovations in the computational field with the potential to positively impact public health and enhance the quality of life of people with neurodegenerative diseases.