Statistical Learning Approaches for Multimodal Alzheimer’s Disease Diagnosis and Progression Prediction: Methods, Challenges, and Future Directions

Alzheimer’s disease (AD) is a neurodegenerative disorder affecting cognitive functions in the elderly. Although no cure exists, early diagnosis and accurate prediction of disease progression are essential for timely intervention. Recently, machine learning (ML) and deep learning (DL) models have been widely applied to this task, leveraging statistical methods to analyze multimodal data such as neuroimaging, electroencephalography (EEG), and fluid biomarkers. This overview examines 128 studies published between 2019 and 2024, focusing on the statistical foundations of ML/DL approaches in AD research, including uncertainty quantification, validation design, feature selection (FS), and the generalizability of predictive models. Particular emphasis is placed on data preprocessing, FS, model validation strategies, and evaluation metrics (e.g., sensitivity, specificity). We also discuss challenges related to imbalanced datasets and the statistical techniques used to address them, providing insights into the reliability and generalizability of ML/DL models for AD diagnosis and prognosis.