Diagnosing Parkinson's disease through artificial intelligence using sound samples

Abstract

 Phonological changes in PD reflect early neurodegenerative changes that affect speech motor control. This paper reviews the development of remote monitoring methods from early acoustic analysis to modern deep learning-based methods. As a non-invasive, low-cost and sensitive tool, speech-based artificial intelligence biomarkers can be used for screening, early diagnosis and longitudinal monitoring. Its advantage lies in its ability to detect subtle changes that are difficult for humans to detect and its potential for high-frequency remote assessment via smartphones. However, some problems remain to be solved, such as the difference in recording environment, the limited annotation data set, and the "black box" nature of AI models limiting the clinical interpretability. In recent years, advances in interpretable AI and multimodal data fusion have brought hope to clinical applications. Future research should focus on improving data diversity, model robustness and transparency, and combine with clinically relevant outcome indicators to promote the wide application of voice AI biomarkers in personalized PD management.