Abstract
Background: Autism spectrum disorder (ASD) and related neurodevelopmental conditions are a significant public health concern, with diagnostic delays hindering timely intervention. Traditional assessments often lead to waiting times exceeding a year. Advances in artificial intelligence (AI) and biomarker-based screening offer objective, efficient alternatives for early identification. Objective: This review synthesizes the latest evidence for AI-enabled technologies aimed at improving early ASD identification. Modalities covered include eye-tracking, acoustic analysis, video- and sensor-based behavioral screening, neuroimaging, molecular/genetic assays, electronic health record prediction, and home-based digital applications or apps. This manuscript critically evaluates their diagnostic accuracy, clinical feasibility, scalability, and implementation hurdles, while highlighting regulatory and ethical considerations. Findings: Across modalities, machine learning approaches demonstrate strong accuracy and specificity in ASD detection. Eye-tracking and voice-acoustic classifiers reliably differentiate for autistic children, while home-video analysis and Electronic Health Record (EHR)-based algorithms show promise for scalable screening. Multimodal integration significantly enhances predictive power. Several tools have received Food and Drug Administration clearance, signaling momentum for wider clinical deployment. Issues persist regarding equity, data privacy, algorithmic bias, and real-world performance. Conclusions: AI-enabled screeners and diagnostic aids have the potential to transform ASD detection and access to early intervention. Integrating these technologies into clinical workflows must safeguard equity, privacy, and clinician oversight. Ongoing longitudinal research and robust regulatory frameworks are essential to ensure these advances benefit diverse populations and deliver meaningful outcomes for children and families.