Artificial Intelligence-Integrated Fmri Methodologies for Early Diagnosis, Prognostic Modeling, and Precision Therapeutic Strategy Development in Alzheimer’S Disease: A Comprehensive Narrative Review Publisher



Mahakizadeh S ; Hassanzadeh G ; Makkiabadi B ; Asgari FHA ; Shafaghi L
Source: Journal of Contemporary Medical Sciences Published:2025

Abstract

Objective: This narrative review examines applications of artificial intelligence to fMRI in Alzheimer’s disease, with emphasis on deeplearning methods. It summarizes how studies approach identification of preclinical functional patterns and assessment of treatment responsiveness, and documents common modeling choices, evaluation practices, and limitations. Methods: This narrative review collates machine-and deep-learning uses of fMRI in Alzheimer’s disease under three themes: Disease-state delineation, therapy-response prediction, and progression modeling. Reported evaluation practices include patient-level site-held-out or external validation with leakage control, calibration (Brier, ECE), decision-curve analysis, scanner harmonization audits, and uncertainty/interpretability assessments. Approaches seen include CNNs, GNNs, and multimodal fusion; robustness strategies include self-supervision, domain-invariant training, and federated learning. Results: Studies have used AI models to characterize patterns in functional connectivity, explore links to clinical trajectories, and examine responder/non-responder distinctions. Multimodal combinations of fMRI with clinical, genetic, or molecular measures often report higher cross-validated performance than fMRI alone, though findings are heterogeneous and sensitive to analytic choices. Persistent constraints include limited interpretability, small and fragmented datasets, and cross-platform variability, which together limit generalizability and clinical applicability. Conclusion: AI-driven analysis of fMRI data shows promise in supporting the study and potential clinical management of Alzheimer’s disease, including early detection and personalized treatment strategies. Future work should focus on improving model interpretability, standardization, and ethical oversight to ensure reliable and responsible application. © 2025, Nab'a Al-Hayat Foundation for Medical Sciences and Health Care. All rights reserved.