Artificial intelligence algorithms for the prediction of cognitive decline and dementia: A theoretical–historical narrative review

Abstract

Population aging has increased the prevalence of neurodegenerative disorders associated with dementia, with Alzheimer's disease remaining the leading cause worldwide. Early identification of individuals at risk of cognitive decline, particularly those with mild cognitive impairment, remains a major clinical and scientific challenge. In this context, artificial intelligence has emerged as a promising framework for prognostic modeling and for the identification of clinically relevant predictors of disease onset and progression. This narrative review provides a theoretical and historical synthesis of the main machine learning and deep learning approaches used to predict cognitive decline and dementia from multimodal data, including biomarkers, clinical and demographic variables, cognitive measures, and neuroimaging markers. The review examines the methodological evolution of the field, from traditional statistical approaches to ensemble methods and deep neural architectures, and discusses the growing role of multimodal and longitudinal data integration. It also highlights major unresolved issues, including limited external validation, reduced interpretability, restricted generalizability, and the gap between predictive performance and clinical implementation. Overall, artificial intelligence has substantially advanced prediction research in cognitive decline and dementia, but its routine clinical adoption will depend on more rigorous validation, improved transparency, and stronger integration between neuroscience, clinical practice, and data science.