Early Detection Methods for Autism Spectrum Disorder: From Clinical Screening to Multimodal AI

Area of Science:

• Neurodevelopmental disorders
• Artificial Intelligence in Medicine
• Biomedical Engineering

Background:

• Early detection of autism spectrum disorder (ASD) is crucial for intervention, but traditional methods can cause delays.
• Current screening relies on parent reports, clinical observation, and specialist assessments.
• There's a growing need for more efficient and objective early identification methods.

Purpose of the Study:

• To map the evolution of early ASD detection methods, from traditional screening to AI-driven approaches.
• To review the methodological landscape and identify emerging trends in early ASD identification.
• To assess the potential and limitations of AI in early ASD detection.

Main Methods:

• A scoping review of literature published between January 2010 and May 2026.
• Searches conducted across biomedical, psychological, and engineering databases.
• Inclusion of 65 evidence sources for qualitative synthesis and methodological guidelines for appraisal.

Main Results:

• Early ASD detection is moving towards multidimensional risk characterization beyond single-session assessments.
• Clinical tools are foundational, complemented by eye tracking, motor analysis, vocal biomarkers, EEG, fNIRS, and genomic data.
• AI methods show promise in quantifying various biomarkers but face challenges like data limitations, validation, bias, and interpretability.

Conclusions:

• Future ASD detection systems should integrate AI as clinician-supervised decision-support tools, not autonomous diagnostics.
• Robust external validation, privacy-preserving deployment, and interpretable AI are essential for clinical progress.
• Aligning AI outputs with developmental and clinical knowledge is key for effective early ASD identification.