Back to all papers

Artificial intelligence applications in fMRI-based disease diagnosis: A systematic review.

October 8, 2026pubmed logopapers

Authors

Almetwally EM,Diab LS,Ahmed MA,Kamel AR

Affiliations (4)

  • Department of Mathematics and Statistics, College of Science, Imam Mohammad Ibn Saud Islamic University (IMSIU), Riyadh, Saudi Arabia.
  • Computer Science Department, Faculty of Computers and Artificial Intelligence, Beni-Suef University, Beni-Suef, Egypt.
  • Department of Applied Statistics and Econometrics, Faculty of Graduate Studies for Statistical Research, Cairo University, Giza, Egypt.
  • Department of Basic Sciences, El-Gazeera High Institute for Computers and Information Systems, Ministry of Higher Education, El-Mokattam, Egypt.

Abstract

The purpose of this study is to critically assess and thoroughly explore the use of artificial intelligence (AI) methods in functional magnetic resonance imaging (fMRI) for the diagnosis of neuropsychiatric and neurological conditions. Methodological variability, dataset features, validation techniques, repeatability transparency, and the clinical preparedness of AI-based diagnostic models are the main topics of the review. A thorough search for research published between 2010 and 2025 was conducted using PubMed, Scopus, Web of Science, and IEEE Xplore. The review was conducted in accordance with PRISMA 2020. The PICOS framework was used to define the eligibility requirements. Studies were included if they published quantitative performance metrics and used resting-state or task-based fMRI data in conjunction with machine learning or deep learning models for disease classification. A modified QUADAS-2 framework appropriate for AI-based diagnostic research was used to evaluate the methodological quality and bias risk. Because study designs, datasets, preprocessing pipelines, and evaluation metrics varied, the data were synthesized qualitatively. After the PRISMA 2020 screening procedure, 80 studies in all satisfied the inclusion requirements. The majority of research concentrated on autistic spectrum disorder, Parkinson's disease, Alzheimer's disease, and major depressive disorder. While machine learning methods (especially support vector machines) and deep learning architectures like convolutional neural networks and graph neural networks were the most commonly used artificial intelligence techniques, resting-state fMRI was the most commonly used imaging modality. In general, reported categorization accuracy fell between 85% and 96%. Nonetheless, significant methodological variation was noted in model topologies, validation techniques, feature extraction techniques, and preprocessing pipelines. Concerns about the generalizability of the model and possible overfitting were highlighted by the frequent dependence on very small cohorts and the sparse use of external validation. Additionally, quality assessment revealed moderate to high risks of bias in several studies, especially in relation to validation and patient selection processes. The use of AI to analyze fMRI data shows great promise for enhancing the diagnostic categorization of neurological and neuropsychiatric disorders. However, clinical translation is restricted by small sample sizes, inconsistent methods, and a lack of external validation. To improve repeatability and facilitate the incorporation of AI-driven fMRI analysis into clinical practice, standardized preprocessing procedures, bigger multi-site datasets, open reporting procedures, and stringent external validation are necessary.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.