One of the most comprehensive studies on face processing in autism has found that neural signals associated with recognising faces are less distinct in autistic children and do not refine with age as they do in neurotypical children. The findings were published in Nature Mental Health.
Researchers at Yale University used electroencephalography (EEG) to measure electrical activity from 128 scalp electrodes in approximately 400 autistic and neurotypical children. Participants viewed images of faces and objects whilst researchers recorded brain activity, then used machine learning to predict what participants were looking at based on the neural signals.
The team found it was easier to determine whether a neurotypical child was looking at a face than whether an autistic child was. Lead author Dr Jason Griffin, now at the University of Houston, explained that neural activity measured at the scalp is simply not as distinct in autistic children.
Furthermore, in neurotypical children, the ability to predict what participants were looking at increased with age, reflecting progressive refinement of face processing. This developmental trajectory was not observed in autistic children.
The research also revealed that differences in face processing appear earlier than previously understood. Most prior research focused on a biomarker occurring 170 milliseconds after seeing a face, but this study found differences arising even earlier in the processing timeline.
Senior author Professor James McPartland of the Yale Child Study Center noted that understanding these developmental differences could inform when interventions might be most effective. The findings suggest differences in face processing stem from underlying neurodevelopmental factors rather than the experience of living with autism.
The researchers suggest this work represents a first step toward introducing new biomarkers that could improve autism diagnosis and identify candidates for supportive interventions.
Source: Medical Xpress / Yale University (Nature Mental Health, 2026)
