Applications of artificial intelligence in the diagnosis and prognosis of Alzheimer’s disease: a systematic review

Abstract

Background: Alzheimer's Disease (AD) is the leading cause of dementia in the elderly, accounting for 60–80% of cases worldwide. It is characterized by progressive memory loss, cognitive decline, and behavioral changes. Care costs exceed $232 billion, while early diagnosis could save up to $7.9 trillion. Definitive diagnosis still requires post-mortem neuropathological confirmation, and available clinical methods have significant limitations, particularly in the pre-clinical phase. In this context, Artificial Intelligence (AI), through Machine Learning (ML) and Deep Learning (DL), has shown promise in integrating large heterogeneous datasets and identifying complex patterns undetectable by traditional methods. Objective: To analyze the use of AI in the diagnosis and prognosis of AD through a systematic literature review, evaluating efficacy, limitations, and clinical viability of ML and DL models. Methods: The research question followed the PICOT framework. The systematic review was conducted according to PRISMA 2020 guidelines, searching MEDLINE (PubMed), Cochrane Library, and SciELO. A total of 2,858 articles were identified, of which 8 were included in the final analysis. Results: DL models outperform ML in complex data analysis. LSTM networks excel in time-series analysis, enabling personalized monitoring of disease progression. CNNs show high performance in MRI analysis, and hybrid CNN-LSTM models achieve 90–91% diagnostic accuracy. A support vector model applied to hippocampal functional connectivity data from 119 individuals reached 82.02% accuracy differentiating AD from healthy controls, 81.33% between MCI and controls, and 81.08% between AD and MCI. A CAM-CNN model achieved 97.35% precision for mild AD, 87.82% for MCI likely to progress, and 78.79% for stable MCI. Biomarkers such as Aβ42, t-Tau, p-Tau, and hippocampal atrophy show 85–95% diagnostic accuracy. Random Forest and Gradient Boosting models proved effective in multimodal plasma analysis. Wearables and sensors now collect digital biomarkers, motor function, speech, cognition, and heart rate variability, with tools like Altoida iADL detecting neurological changes before clinical symptoms appear. Conclusion: AI advances, particularly DL and ML, show strong potential for early AD detection. Models such as LSTM, CNN, and digital biomarker tools enable more accurate, non-invasive diagnoses. Standardization, external validation, and interpretable AI remain priorities for future research.