Artificial Intelligence for Early Alzheimer’s Disease Diagnosis and Progression Analysis: A Comprehensive Survey of Deep Learning, Multimodal Learning, Graph Networks, and 3D Transformers

Abstract

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder in which early and reliable diagnosis remains challenging because pathological changes often begin years before severe clinical symptoms become evident. Recent advances in artificial intelligence (AI) have enabled increasingly sophisticated analysis of neuroimaging, cognitive, genetic, and biomarker data for AD diagnosis, staging, and progression prediction. This survey presents a comprehensive review of AI-based approaches for Alzheimer’s disease, covering conventional machine-learning methods, 2D and 3D convolutional neural networks, multimodal deep learning, contrastive and metric learning, graph neural networks, Siamese architectures, and Transformer-based models. Particular attention is given to the use of structural MRI, fMRI, PET, clinical and cognitive variables, genetic information, and fluid biomarkers, together with commonly adopted preprocessing, harmonization, and multimodal fusion strategies. The survey also examines explainability, uncertainty estimation, cross-dataset robustness, longitudinal progression modelling, computational efficiency, and clinical generalizability. Comparative analysis of representative studies indicates that while very high performance is frequently reported for AD-versus-cognitively-normal classification, more clinically relevant tasks such as mild cognitive impairment conversion and individualized progression prediction remain substantially more difficult. To address these limitations, a conceptual future architecture is proposed that integrates modality-specific foundation encoders, cross-modal contrastive alignment, a structure-aware multimodal Transformer, dynamic patient-graph reasoning, longitudinal temporal modelling, multi-task prediction, and an explainable clinical layer. Overall, the survey highlights the need to move beyond single-dataset accuracy toward robust, interpretable, uncertainty-aware, longitudinal, and clinically deployable AI systems for early Alzheimer’s disease diagnosis and progression assessment.