The research, led by Kacie Dunham-Carr and colleagues at Vanderbilt University Medical Center together with Micah Murray’s team in Lausanne, Switzerland, revisited electroencephalography data from fifty school-aged children, twenty-five autistic and twenty-five non-autistic, matched at the group level on chronological age and sex assigned at birth. The children, aged roughly five and a half to twelve and a half years, watched two-second videos of an adult female speaker pronouncing the consonant-vowel syllable “ba.” In the auditory-only condition, the syllable was paired with a still image of the speaker’s face; in the audiovisual condition, the sound was synchronized with the natural movements of the face and neck. Each child completed one hundred trials of each stimulus type while a 128-channel sensor net recorded their brain activity at one thousand samples per second, with an eye tracker ensuring that children were actually looking at the fixation cross before each trial began.

What makes this study methodologically distinctive is its departure from the conventional playbook of event-related potential analysis. Traditional approaches measure the amplitude and latency of recognizable peaks and troughs, such as the P2 component that appears roughly two hundred milliseconds after stimulus onset, at one or a handful of experimenter-selected electrodes. That strategy carries well-known liabilities. The measured voltage at any electrode depends entirely on the choice of reference electrode, and statistical effects can change their pattern or vanish altogether when a different reference is chosen, with no empirical basis for preferring one over another. Analyses at a single midline electrode, such as the vertex site Cz used in many prior studies, also discard most of the information collected by high-density recordings and cannot distinguish whether an observed effect reflects a change in response strength, in the spatial configuration of underlying brain sources, or in timing.

To sidestep these pitfalls, the team turned to reference-independent topographic metrics computed across the entire electrode montage. The first, global field power, quantifies the overall strength of the electric field at each instant by calculating the spatial standard deviation of voltage across all electrodes. The second, global map dissimilarity, measures how different the spatial distribution of the electric field is between two responses, independent of their strength. Because biophysics dictates that changes in scalp topography can only arise from changes in the configuration of intracranial neural sources, global map dissimilarity offers a data-driven window into whether different groups or conditions engage different brain networks, even before any source modeling is attempted. The researchers computed these metrics on difference waveforms, created by subtracting each child’s auditory-only response from their audiovisual response, thereby isolating the interaction between group and condition.

The strength results were strikingly consistent with earlier conventional findings. Global field power was significantly reduced for audiovisual compared to auditory-only speech between 245 and 275 milliseconds after stimulus onset, a window overlapping the P2 component, and this facilitation effect appeared in both autistic and non-autistic children. When the team compared the strength of the difference waveforms between groups, they found no significant differences, confirming that the magnitude of the audiovisual benefit, as indexed by response suppression, is essentially the same in both populations. This convergence across conventional and reference-independent strength measures suggests that autistic children derive a comparable reduction in neural response when visual speech cues accompany auditory speech, a pattern the authors interpret as a facilitation effect of successful multisensory integration operating in both groups.

The topography told a different story. Global map dissimilarity analyses, evaluated with non-parametric permutation statistics known as topographic ANOVA, revealed significant group differences in the spatial configuration of the difference waveforms between 139 and 165 milliseconds post-stimulus, a timeframe comparable to the P2 window implicated in phonetic processing. To characterize these differences, the researchers applied a hierarchical clustering algorithm to the group-averaged data, which identified stable template maps, or microstates, that could then be fit back to each individual child’s data through spatial correlation. Over the critical window, non-autistic children’s differential response was best characterized by a template map showing a left-lateralized positive distribution over occipito-temporal scalp regions, whereas autistic children’s differential response was dominated by a map with a bilateral positive distribution over occipital sites. A mixed-model analysis of the time spent in each map confirmed a significant group-by-map interaction, with a substantial effect size.

Source localization analyses reinforced this topographic dissociation. Using a distributed linear inverse solution with anatomical constraints, the team estimated the intracranial generators of the difference waveform over the 139 to 170 millisecond window and compared the estimates between groups with randomization tests. Autistic children showed greater estimated source activity than their non-autistic peers in left parietal cortex, the left middle temporal gyrus, and the left inferior temporal regions, alongside reduced estimated activity in the right middle temporal gyrus, the anterior cingulate cortex, and the cuneus. The occipital emphasis in the autistic group’s estimated sources dovetails with a prior functional MRI study from the same laboratory, which found that autistic youth activated predominantly primary visual and motor cortices in response to audiovisual speech, fewer regions than the canonical auditory and visual cortices and posterior superior temporal cortex engaged by non-autistic adults.

Taken together, the findings point to a provocative interpretation: autistic and non-autistic children may arrive at superficially similar scalp responses through different perceptual routes. Because the voltage waveform recorded at the vertex electrode can be produced by multiple combinations of underlying sources, two groups with genuinely distinct neural networks for combining lip movements with speech sounds can nonetheless yield indistinguishable amplitudes at a single electrode. The authors suggest that a more visually dominant processing route in autistic children could produce the bilateral occipital topography observed here, while non-autistic children recruit a left-lateralized occipito-temporal network more consistent with classical audiovisual speech regions. Notably, the group differences emerged during early-stage sensory processing windows rather than later intervals associated with higher-level cognitive and linguistic operations, likely because the simple syllable stimuli used were designed to isolate early integration rather than semantic processing.

The study has limitations that the authors acknowledge candidly. The sample consisted of children with spoken language who could tolerate wearing an EEG cap and following simple directions, and was relatively homogeneous in race and ethnicity, so generalization to the broader autism spectrum remains to be demonstrated. The single-syllable stimuli, while ideal for probing early integration, may not capture the complexities of natural connected speech, and the statistical thresholds for significance in ERP time series, though grounded in foundational work on temporal autocorrelation, retain an element of arbitrariness that remains an active area of methodological research. Even so, the paradigm is brief, passive, and relatively noninvasive, and the reference-free metrics it showcases can be compared directly across future studies in ways that conventional peak-amplitude analyses cannot.

The broader implication is that autism research may need to look beyond the amplitude of brain responses to find where sensory processing truly diverges. Differences in signal processing, the authors conclude, may not always be apparent through differences in ERP amplitudes, and analyses that treat the full scalp distribution of electrical activity as the object of study can reveal neurophysiological distinctions invisible to component-based approaches. For a field that has long grappled with mixed neuroimaging results despite robust behavioral evidence of altered audiovisual speech perception in autism, the demonstration that response strength and response topography can dissociate offers both an explanation for past inconsistencies and a practical roadmap for future investigations of the neural circuits underlying multisensory development.

Subject of Research: Audiovisual speech processing differences in autistic versus non-autistic children measured with topographic event-related potential analyses

Article Title: Scalp Topography of Event-Related Potentials Reveals Differences in Audiovisual Speech Processing in Autistic and Non-Autistic Children

Article References: Scalp Topography of Event-Related Potentials Reveals Differences in Audiovisual Speech Processing in Autistic and Non-Autistic Children. (n.d.). https://doi.org/10.1007/s10803-026-07552-5

Image Credits: AI Generated

DOI: 10.1007/s10803-026-07552-5

Keywords: autism, EEG, event-related potentials, audiovisual speech, multisensory integration, global field power, global map dissimilarity, topographic analysis, sensory processing, children, source localization, P2 component