Artificial intelligence-driven neuroimaging for early diagnosis of Alzheimer’s disease: a systematic review

Abstract

Background: Alzheimer’s disease is the leading cause of dementia worldwide, affecting approximately 50 million individuals. Early diagnosis remains a major challenge, as conventional neuroimaging methods often detect structural changes only after significant neurodegeneration has occurred. Advances in artificial intelligence have enabled computational models capable of identifying subtle neuroimaging patterns associated with early disease stages. Machine learning and deep learning techniques have increasingly been applied to neuroimaging data to detect early structural and metabolic alterations, potentially improving diagnostic accuracy and enabling earlier clinical intervention. Objective: To evaluate the diagnostic performance of artificial intelligence models applied to neuroimaging for the detection and early diagnosis of Alzheimer’s disease. Methods: A systematic review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines using a search strategy structured according to the Population, Intervention, Comparison and Outcome framework. The protocol was prospectively registered in the International Prospective Register of Systematic Reviews. Studies investigating artificial intelligence algorithms, including machine learning and deep learning approaches applied to neuroimaging for diagnosing Alzheimer’s disease or mild cognitive impairment, were included. Searches were performed in PubMed, Cochrane Library and LILACS for studies published between 2021 and 2026. A total of 2,085 records were identified. After filtering, 49 studies were screened using the Rayyan platform. Following duplicate removal and title and abstract screening, 12 articles underwent full-text evaluation and 8 met the eligibility criteria for qualitative synthesis. Results: The included studies applied several artificial intelligence techniques, including convolutional neural networks, generative adversarial networks, random forest models and other deep learning architectures. Neuroimaging modalities included structural magnetic resonance imaging, functional magnetic resonance imaging and positron emission tomography, including amyloid and tau imaging techniques. Diagnostic performance was high, with accuracy ranging from approximately 85% to 94% and area under the receiver operating characteristic curve between 0.87 and 0.94. Multimodal approaches integrating different imaging modalities or biomarkers showed superior performance compared with single-modality models. The hippocampus, medial temporal cortex and temporoparietal regions were the brain areas most frequently associated with diagnostic prediction. Conclusion: Artificial intelligence applied to neuroimaging shows strong potential to improve the early diagnosis of Alzheimer’s disease, demonstrating high diagnostic accuracy and the ability to identify relevant neurodegenerative patterns that may support clinical decision-making in neurodegenerative disorders.