Machine learning and deep learning in clinical practice: Advancing neurodegenerative disease diagnosis with multimodal markers

Neurodegenerative diseases, including Alzheimer's disease, Parkinson's disease, Huntington's disease, and multiple sclerosis, present major global health challenges due to their progressive and incurable nature. Early and accurate diagnosis is critical to slow disease progression and optimize therapeutic interventions, yet conventional diagnostic approaches, such as neuroimaging, cerebrospinal fluid biomarker analysis, and clinical evaluation are often inadequate at the prodromal stage. Recent advances in artificial intelligence, particularly machine learning (ML), have provided new opportunities for precision diagnosis and treatment in neurology, using large data and multimodal biomarkers. Applications of ML to data from neuroimaging, electrophysiology, behavioral functions, speech and handwriting analysis, and molecular biomarkers have shown promising improvements in diagnostic accuracy, patient classification, and therapeutic recommendations. However, significant challenges remain, including data heterogeneity, model interpretability, population diversity, and ethical concerns surrounding patients' privacy. The purpose of this review is to examine current applications of ML in the diagnosis and management of neurodegenerative diseases through various data, highlight its strengths and limitations, and discuss future directions for using these approaches in clinical practice. We also outline emerging directions, including multimodal fusion with longitudinal data, federated and privacy-preserving learning, and the potential of explainable AI (XAI) and large language models (LLMs) in clinical decision support.