Prediction and Diagnosis for Autism Spectrum Disorder

Autism spectrum disorder (ASD) is a life-long, heritable neurodevelopmental condition affecting over 1% of the global population. Individuals with ASD exhibit significant heterogeneity in clinical phenotypes yet consistently manifest deficits in social communication and restricted repetitive behaviors. Despite its prevalence and clinical significance, validated intervention or treatment protocols for ASD remain scarce. The considerable positive impact of early intervention suggests the imperative of prioritizing early diagnosis. Magnetic resonance imaging (MRI) has enabled the localization of brain alterations associated with specific behaviors. This has provided valuable insights into the neurobiological foundations of ASD while also identifying targets for early diagnosis and personalized interventions. , Several MRI-based biomarkers and predictive models have been recently introduced, demonstrating promising efficacy in identifying autistic symptoms from infant brain scans. These findings are fascinating as they suggest the feasibility of early ASD prediction. In light of these advances, we present a comprehensive literature review focusing on predictive modeling in ASD, specifically focusing on studies using multimodal MRI data. Our review delves into recent strides made in delineating potential early structural, functional, and connectomic signatures of ASD, machine learning methodologies for case-control classification, early diagnosis, symptom prediction, and the utility of neurobiological subtyping in disentangling ASD heterogeneity. Emphasis is placed on elucidating how these methods can enhance our understanding of the complex mechanisms underlying ASD and their translational implications for clinical research and practice. We conclude by considering future directions for advancing predictive modeling in ASD.