Artificial Intelligence for MRI-Based Identification of Autism Spectrum Disorder: A Systematic Review of Methods, Performance, and Clinical Translation

Abstract

Background Autism Spectrum Disorder (ASD) is a neurodevelopmental condition characterized by impairments in social communication and restricted or repetitive behaviours. Early diagnosis is important for timely intervention, motivating increasing interest in computer-aided diagnostic systems based on neuroimaging data. Objective This review aims to systematically review MRI-based machine learning and deep learning studies that classified individuals with ASD and controls and to evaluate current dataset, preprocessing methods, modelling approaches, validation strategies, and barriers to clinical translation. Methods A systematic literature review was conducted following the PRISMA guidelines. Studies published between 2015 and 2026 were retrieved from Scopus, Web of Science, and PubMed databases. After applying predefined inclusion and exclusion criteria, 86 journal articles were selected for detailed analysis. Results The reviewed studies were categorized according to the imaging modality employed, including structural MRI (sMRI), functional MRI (fMRI), and multimodal approaches. The analysis covered datasets, preprocessing pipelines, feature engineering techniques, brain atlases, and classification models. fMRI was the most frequently used modality, while graph neural networks, transformers, and multimodal fusion frameworks emerged as the dominant methodological trends. Despite encouraging results, challenges related to dataset heterogeneity, limited sample size, model interpretability, and clinical generalizability remain. Conclusions MRI-based machine learning and deep learning methods show considerable potential for ASD diagnosis. Future research should focus on multimodal integration, explainable AI, and large-scale harmonized datasets to support clinically reliable diagnostic systems.