A Comprehensive Literature Review on Multimodal Explainable AI Techniques for Early Detection of Parkinson’s Disease

Parkinson's Disease (PD) is a neurodegenerative disease characterized by its progressive nature and difficulties associated with the early diagnosis due to subjective neurological examinations and motor symptom presentation only during later stages of the disease. Advances in the fields of Artificial Intelligence (AI), Machine Learning (ML) and Deep Learning (DL) now allow for automated and objective detection of PD through analysis of biomedical signals, specifically speech, handwriting, and gait measurements. This paper offers an overview of the existing AI-driven multimodal approaches used for early PD diagnosis paying special attention to the role played by explainability methods in creating reliable and transparent algorithms. The recent scientific developments in multimodal analysis of speech, handwriting dynamics, and gait as well as feature fusion strategies and interpretable deep learning architectures are analyzed. CNNs, RNNs, Transformer Models, Siamese Networks and various self-supervised methods are considered in detail. Moreover, the increasing use of multimodal systems which integrate different biomedical data sources into one model is highlighted. Research gaps in this domain include the lack of practical validation of approaches in real-world conditions, vulnerability to absence of some modality, high complexity, potential data bias and lack of inherently interpretable models. Finally, perspectives for future studies in the area of multimodal AI-driven PD detection are discussed.