A Review on Early-Stage Risk Prediction for Alzheimer's Disease

Early prediction of Alzheimer’s disease (AD) prior to irreversible neurodegeneration is a critical global healthcare priority. This paper presents a PRISMA-based systematic review of machine learning (ML) and deep learning (DL) approaches for forecasting the conversion from Mild Cognitive Impairment (MCI) to AD. Analyzing the last decade (2016–2026), we highlight a distinct shift from unimodal, handcrafted feature models to fully automated, multimodal architectures. Specifically focusing on the last three years, this review categorizes state-of-the-art works by their data modalities—ranging from neuroimaging and blood biomarkers to digital speech and electroencephalography (EEG). We detail the mathematical mechanisms of multimodal fusion, evaluate major recent works, identify current research gaps (such as clinical generalizability and the "black-box" dilemma), and discuss future trajectories.