Artificial Intelligence for Predicting Mild Cognitive Impairment to Alzheimer’s Disease Conversion: A Systematic Review of Advances, Challenges, and Clinical Implications

Abstract Alzheimer’s disease (AD) is a rising global health concern, with mild cognitive impairment (MCI) recognized as a transitional stage. Timely prediction of MCI-to-AD conversion allows for earlier clinical interventions, yet traditional methods lack sensitivity. Artificial intelligence (AI) models offer improved accuracy by utilizing neuroimaging, biomarkers, and cognitive assessments. This systematic review aimed to evaluate recent advancements in AI-based approaches for predicting MCI-to-AD conversion, identify key methodological trends, highlight clinical implications, and discuss limitations and future directions. Preferred Reporting Items for Systematic Reviews and Meta-Analyses standards were followed in conducting a systematic review. Studies published between January 2020 and March 2025 were retrieved from PubMed and Google Scholar. Inclusion criteria were peer-reviewed articles using AI for MCI-to-AD prediction involving neuroimaging, biomarkers, or multimodal frameworks, and reporting accuracy, area under the curve (AUC), sensitivity, or specificity. Interpretability and prediction performance were among the results evaluated. The QUADAS-2 technique was used to evaluate the quality of the study. Twenty studies met the inclusion criteria. Deep learning models, especially convolutional neural networks, consistently outperformed traditional machine learning approaches, achieving AUCs between 0.85 and 0.92. Multimodal AI models integrating magnetic resonance imaging, positron emission tomography, biomarkers, and cognitive data demonstrated higher predictive power. The use of explainable AI methods like Grad-CAM and Shapley additive explanations increased transparency. However, many studies were limited by small sample sizes and a lack of external validation, affecting generalizability. AI has a lot of potential for early AD diagnosis and prognosis. Yet, challenges remain in data standardization, model bias, and clinical implementation. Future research should focus on multicenter validation, regulatory compliance, and clinician training to enable integration into routine dementia care.