A Scoping Review of AI-Based Approaches for Detecting Autism Traits Using Voice and Behavioral Data.

Area of Science:

• Computational neuroscience
• Developmental psychology
• Biomedical informatics

Background:

• Autism Spectrum Disorder (ASD) diagnosis relies on behavioral observation, crucial for early intervention.
• Computational behavioral phenotyping using Artificial Intelligence (AI) is rapidly advancing for ASD diagnostics.

Purpose of the Study:

• To systematically map the application of AI in Autism Spectrum Disorder diagnostics.
• To focus on computational behavioral phenotyping across key behavioral modalities.

Main Methods:

• Scoping review of 158 studies published between 2015 and 2025.
• Analysis of AI applications in eight behavioral modalities: voice, conversation, linguistics, movement, activity, facial gestures, visual attention, and multimodal approaches.
• Synthesis of Machine Learning and Deep Learning techniques in ASD diagnosis.

Main Results:

• AI demonstrates high diagnostic performance (up to 99% accuracy) in controlled settings.
• Identified significant challenges: dataset heterogeneity, gender bias, and small sample sizes.
• Comprehensive overview of AI's current strengths and weaknesses in ASD diagnosis.

Conclusions:

• AI holds significant promise for ASD diagnostics, particularly with advanced ML/DL techniques.
• Clinical implementation and generalizability are hindered by data limitations and biases.
• Future research should focus on inclusive, generalizable, and clinically applicable AI diagnostic solutions.