Predicting, Detecting, and Monitoring Cognitive Impairments using Artificial Intelligence: A Systematic Scoping Review

Background and aims: Due to an aging population and changing lifestyles, cognitive declineis a controversial research topic. Also, advances in artificial intelligence (AI) are being used toimprove healthcare, including monitoring and diagnosing cognitive impairment. In this study, wedecided to explore the global use of AI for monitoring and diagnosing cognitive disorders througha scoping systematic review.Method: Using the PRISMA guidelines, we searched comprehensively in PubMed without languageor time restrictions. Two researchers independently reviewed the articles based on titles andabstracts and finally extracted relevant data from the included articles.Results: Initial records identified through PubMed searching included ۳۵۵ articles. Accordingly,۱۶۹ met the eligibility criteria and were included in data extraction.۱۲۹ studies (۷۶%) had beenperformed in the three last years. The USA and South Korea had the most studies (۱۷.۶% and۹.۵%, respectively).Mild cognitive impairments and Alzheimer’s were the top diseases surveyed (both at ۵۵%). Apartfrom the significant volume of studies that focused on neurological and mental diseases, four studiesdealt with cognitive disorders of internal diseases (Diabetes Mellitus and breast cancer (twostudies for each topic). Among studies, assessment of six cognitive domains varied including,memory: ۱۳۵ (۸۰%), attention: ۱۲۴ (۷۳%), language: ۱۲۱ (۷۲%), executive function: ۶۵ (۳۸%),perceptual-motor function: ۴۹ (۲۹%), and social cognition: ۱۲ (۷%). Evaluation of awareness andbehavior were assessed in ۱۰۰ (۵۹%) and ۲۵ (۱۵%) studies, respectively.۹۹ (۵۹%) studies wereconducted with the purpose of diagnosis, while ۳۵ (۲۱%) studies were directed for prediction,and ۳۴ (۲۰%) studies were conducted for monitoring or classification. The most used assessmentquestionnaire for AI data were A Mini-Mental State Examination (MMSE) (۱۰۶), Montreal CognitiveAssessment (MoCA) (۴۱), and Clinical Dementia Rating Scale (۴۰). However, some neuroimagingtools were commonly used, including magnetic resonance image (MRI) and positronemission tomography (PET) (۹۱ and ۲۴ studies, respectively). Among the AI approaches, SupportVector Machine (۸۶), one of the neural network methods (۵۱), Random Survival Forest (۵۳), andLogistic Regression (۴۱) were used more than other algorithms. Cross-validation of AI was donevia five methods including, K-fold (۱۲۳), Leave-one-out (۱۶), Stratified K-fold (۳), Monte Carlo(۲), and Holdout (۱).Conclusion: AI-based prediction, diagnosis, and monitoring of cognitive impairment is a growingfield that has received more attention in recent years. AI can potentially assist both neuropsychologicaland internal diseases with cognitive impairment. Combining various AI methodsyields better results, with Support Vector Machine, Random Survival Forest, and LogisticRegression commonly used among machine learning algorithms. In general, the more complexmodels combined with multimodal data (clinical, cognitive, and neuroimaging) achieved the bestperformance. It is critical to resolving the ambiguities in future studies.

کلیدواژه ها:

Cognitive Impairments ، Artificial Intelligence ، Machine Learning