EEG and Machine Learning Reveal How Autistic Brains Process Faces Differently, Pointing to New Biomarker
核心洞察
A large-scale EEG study of approximately 400 autistic children found that face-related neural signals are less distinct compared to neurotypical peers, published in Nature Mental Health.
Machine learning analysis of whole-scalp electrical activity showed that neurotypical children's face processing becomes more refined with age, a developmental trajectory absent in autistic children.
The findings suggest face processing differences in autism (搜索) emerge earlier than previously understood, beyond the well-known N170 (搜索) biomarker, and may reflect the condition's underlying biology rather than lived experience.
A new study published in Nature Mental Health has provided the most comprehensive view yet of how autistic children process faces, using electroencephalography (EEG) across the entire scalp combined with machine learning. The research, led by James McPartland, PhD, Harris Professor in the Child Study Center and director of the Center for Brain and Mind Health at Yale, found that face-related neural signals are less distinct in autistic children and do not follow the same developmental refinement seen in their neurotypical peers.
"This is the first time we've leveraged the full volume of very rich information provided by an EEG recording to understand face perception and autism (搜索)," McPartland said.
Whole-scalp analysis captures broader differences
The study drew on participant data from the Autism (搜索) Biomarkers Consortium for Clinical Trials (ABC-CT), examining approximately 400 autistic and neurotypical children across a range of ages. While most prior EEG research on face processing has focused on selected brain regions—analogous, McPartland explained, to estimating city traffic with a single downtown camera—the new approach analyzed electrical signals from all 128 electrodes placed around the scalp.
Researchers used machine learning to determine whether patterns in brain activity could predict what each participant was viewing: faces or objects. The system identified when a neurotypical child was looking at a face more readily than when an autistic child was. "The neural activity that we measure at the scalp is just not as distinct in autistic children," said first author Jason Griffin, PhD, assistant professor of psychology at the University of Houston and former postdoctoral researcher in McPartland's lab.
Developmental trajectory diverges
Among neurotypical children, the ability to predict what participants were viewing increased with age, reflecting a refinement and specialization of face processing over time. In autistic children, no comparable age-related change was observed. "Their face-specific processing is not following the same trajectory," Griffin said.
The findings also broaden understanding of when face processing differences arise. Much prior research has centered on the N170 (搜索), a shift in brain activity occurring approximately 170 milliseconds after seeing a face, which is known to be delayed in autistic individuals. The new whole-scalp approach, however, revealed that differences in autism (搜索) occur even earlier in the temporal sequence. "When you look at face processing through an approach that's more inclusive of the full temporal range of information, we're seeing that differences in autism occur even earlier," McPartland said.
Implications for intervention and diagnosis
The study addresses a persistent challenge in autism (搜索) research: distinguishing biological features of the condition from changes shaped by the experience of living with autism. The findings suggest that differences in face processing reflect autism itself rather than later experiences. "When we're seeing this expected refinement and specialization in non-autistic people, we're seeing a reduced developmental change in autistic people," McPartland explained, indicating that interventions aimed at helping autistic people interpret facial emotions might be most effective in earlier childhood, when these developmental differences first emerge.
Many neurodivergent people are not limited by autism (搜索) and may neither need nor want intervention. For others, however, difficulty interpreting faces can significantly affect quality of life. "Imagine navigating life when it's hard for you to perceive the information in another person's face, from navigating a playground to navigating a job interview," McPartland said. "This is one of the critical areas that causes difficulty for many autistic people."
Currently, clinicians diagnose autism (搜索) based on behavior, as no biologically-based assessment tools exist. Researchers are evaluating various biomarkers as candidates for improved diagnostic methods or to define subgroups. The whole-scalp EEG patterns identified in this study may represent a step toward such tools. "This is a first step in introducing a new biomarker," McPartland said.
The study was supported by the National Institutes of Health (awards 40MH134303, U19MH108206, and K01MH137401), Yale University (搜索), and the Hilibrand Foundation.