Use of artificial intelligence in predicting cognitive decline and dementia: a systematic review

Résumé

The use of Artificial Intelligence (AI) and Machine Learning (ML) has been applied in several studies to detect the conversion from mild cognitive impairment (MCI) to dementia. This study aims to identify challenges and effective methods for applying AI in predicting cognitive decline and dementia. A literature review was conducted by searching the PubMed database between May and July 2024, using predefined eligibility criteria. The analysis included 17 studies involving 12,183 participants (aged 50 to 85 years). The review included 17 articles published between 2015 and 2023, predominantly observational studies. The AI tools applied encompassed demographic and lifestyle data, imaging exams, cognitive tests, biomarkers, and physiological data. MCI was the most prevalent diagnosis among the studies, with a particular focus on the amnestic subtype. Nine studies investigated the conversion from MCI to some form of dementia; four analyzed the prediction of beta-amyloid positivity; three employed ML for diagnostic support in distinguishing controls from patients with different types of dementia; and one assessed the cognitive risk in an elderly population. Among the best-performing approaches identified, multimodal data strategies — especially those involving MRI imaging and sociodemographic information — stood out. Applications involving multimodal datasets, particularly those including MRI exams, improve the overall performance of ML models and are considered among the most effective approaches. Future research should focus on the diagnosis and prediction of less-studied dementias, such as frontotemporal dementia, through the use of AI. The integration of multimodal data, including demographic information, imaging exams, cognitive assessments, and biomarkers, should be encouraged.