AI Algorithms Spot Signs of Autism in Faces and Voices with High Precision

Diagnosing autism spectrum disorder (ASD) has traditionally relied on the expert gaze of trained professionals to observe social behaviour. While this remains the clinical gold standard, the process is often time-consuming and prone to subjective interpretation. A new scoping review suggests that artificial intelligence (AI) offers a powerful way to augment these methods by providing objective analysis of facial expressions, voice, and text.

The study highlights the impressive capabilities of deep learning—a type of advanced AI modelled on neural networks—particularly in facial image analysis. Hybrid algorithms combining different machine learning architectures achieved diagnostic accuracy as high as 99 per cent. Voice analysis also proved robust, with systems successfully detecting atypical speech patterns and prosodic abnormalities (variations in rhythm and tone) with accuracy ranging from 70 to 98 per cent.

Text-based analysis using natural language processing also showed promise in identifying linguistic markers associated with ASD. Ultimately, these tools are designed to complement, not replace, clinicians. By offering rapid, data-driven insights, AI could facilitate earlier interventions and improved outcomes for individuals on the spectrum.