Abstract

Early detection of Alzheimer’s disease (AD) enables timely intervention, better patient management, and improved outcomes. This paper reviews recent methods for early AD detection, proposes a multimodal machine-learning framework combining structural MRI, resting-state fMRI, cognitive scores and plasma biomarkers, and evaluates the approach on a benchmark dataset. Results show that multimodal fusion with a lightweight 3D-CNN + transformer attention module improves classification of healthy controls, mild cognitive impairment (MCI) and AD versus single-modality baselines, with higher sensitivity to early (MCI → AD) converters. The study highlights trade-offs between accuracy, interpretability, and clinical feasibility and outlines directions for translation to clinical practice.