Artificial Intelligence for Detection of Parkinson's Disease From Speech Signals-A Comprehensive Review

After Alzheimer's disease, Parkinson's disease (PD) is the second most common neuropathological condition. It is a progressive degenerative disease at superannuation that affects the central nervous system (CNS) and slowly disable patients doing regular activities like walking, speaking, and writing. Early diagnosis of this disease helps to manage the patients and provide them therapy effectively. From the past few years, gait, electroencephalogram (EEG) signals and speech signals have been inspected to detect this disease at an early stage, out of which the most frequently considered one is speech signal, as it is reported by the researchers that 90% of the PD patients suffer from speech disorders. Also, speech signal analysis is a non-invasive and cost-effective method to detect PD at an early stage, and it helps to build telediagnosis models for prediction. Classical speech signal processing methodologies adopted in PD detection sometimes suffer from inadequate understanding of the effect of PD speech generation models and how that is reflected on speech signals captured from the PD patients. Artificial intelligence (AI) based methods attempts to learn those models from the given data in the best possible way to distinguish PD patients from the healthy controls. This paper's primary goal is to survey AI methodologies to detect PD using speech signals as reported in the publications between 2020 and 2024. As deep learning (DL) is a subset of machine learning (ML) and ML is a subset of AI, we consider 55 research publications related to ML and DL methods adopted for speech signal-based PD diagnosis. All the articles were published by IEEE and we have considered key words like "Machine learning approaches in Parkinson's disease detection from speech signals," "Application of Deep learning in Parkinson Disease detection from speech signals," "Artificial Intelligence in Parkinson's disease detection"; to find the articles reviewed in this study. This comprehensive review article reveals that both ML and DL algorithms have demonstrated encouraging outcomes, and the need of focused effort on more explainable AI based methods that can be clinically interpreted and hence potentially can be trusted for early diagnosis of Parkinson's disease.