Live face-to-face tests reveal atypical brain responses in early psychosis

People in the earliest stages of schizophrenia spectrum disorders show atypical brain activity and reduced involuntary social responses when they look at a real person’s face, according to a study that used a two-person live gaze task rather than the static images or video clips common in social neuroscience. Researchers at Yale University recorded brain signals, eye movements and facial muscle activity simultaneously while 24 individuals with first episode psychosis and 32 typically developed participants sat face to face with a lab partner whose natural expressions served as the stimuli. The findings, published in Frontiers in Neuroimaging, point to a specific cortical network centered on the right temporoparietal region that behaves differently in early psychosis.

The paradigm was designed to capture something that conventional social cognition experiments often miss: the reciprocal, immersive quality of a genuine encounter. A lab partner and a participant sat 140 centimeters apart, separated by a programmable smart glass window that switched between opaque and transparent states. During task blocks, the lab partner watched short, emotionally evocative video clips for four seconds and then gazed at the participant’s face for five seconds. Any expression on the lab partner’s face was a genuine, unrehearsed response to the clip content, whether cute animal antics of positive valence or spiders and decaying matter of negative valence. Each run lasted three minutes, alternating 18-second task blocks with 12-second rest blocks, and there were four runs in total, two for each emotional valence.

While the participant observed the live face, the research team collected functional near infrared spectroscopy, or fNIRS, signals from 134 optical channels covering the head, along with 32 channels of electroencephalography at 256 hertz. Eye tracking at 120 hertz recorded gaze fixations and saccades, and scene cameras captured facial movements that were quantified into 17 anatomical action units using OpenFace 2.0 software. The fNIRS technique uses near infrared light at wavelengths between 680 and 900 nanometers to detect changes in oxyhemoglobin and deoxyhemoglobin concentrations, exploiting the neurovascular coupling that makes an active brain region draw more oxygenated blood. Its spatial resolution of one to three centimeters is coarser than that of functional MRI, but its higher temporal resolution and tolerance for natural sitting motion make it well suited to dynamic two-person paradigms that scanners cannot easily accommodate.

The hemodynamic analysis focused on six predefined regions of interest previously linked to live face processing: the right supramarginal gyrus, the right angular gyrus, the right somatosensory association cortex in the superior parietal lobule, the right dorsolateral prefrontal cortex, and the face areas of the right somatosensory and motor cortices. Fifty of the 56 participants contributed usable fNIRS data. Compared with typically developed participants, the psychosis group showed reduced activity in the right supramarginal gyrus, which reached a T value of 2.90 with an adjusted p value of 0.009, and in the face areas of the right somatosensory and motor cortices. Conversely, the psychosis group showed higher activation in the right angular gyrus, the right somatosensory association cortex and the right dorsolateral prefrontal cortex. All comparisons survived false discovery rate correction across the region-of-interest family of tests.

The right supramarginal gyrus holds particular significance because prior work from the same laboratory identified it as specialized for live face gaze. It responds more strongly when people gaze at live faces than at faces shown over video calls or on humanoid robots, and it has been found to be underactive in autism spectrum disorder, another condition marked by social cognition difficulties. The researchers also examined how the participants’ own facial action units, summarized by the first principal component, related to their hemodynamic responses, and the group differences emerged when this behavioral regressor was included in the general linear model analysis.

Exploratory correlation analyses connected the neural differences to clinical symptoms. The expressive deficit score, a composite of four PANSS negative symptom items covering blunted affect, poor rapport, lack of spontaneity in conversation and motor retardation, was negatively correlated with activity in the right supramarginal gyrus and the right dorsolateral prefrontal cortex. Separately, the PANSS item N4, which measures passive and apathetic social withdrawal, was negatively correlated with activity in the right superior parietal lobule, corresponding to the somatosensory association cortex. The authors emphasize that these correlations are exploratory and do not establish causation, but they suggest that reduced activity in core live face processing areas accompanies the negative symptoms that often resist antipsychotic medication.

The EEG recordings added a temporal dimension to the spatial findings. Event related potentials, averaged across a 780-millisecond window after the smart glass opened, showed clear group differences in amplitude and latency within the first 100 to 300 milliseconds, peaking at 200 milliseconds. Five of the seven predefined right temporoparietal electrodes, namely CP2, P8, P4, CP6 and PO4, showed statistically significant differences after false discovery rate correction, with corrected p values ranging from 0.00003 to 0.009. The psychosis group also exhibited delayed latencies of roughly 90 to 100 milliseconds in bilateral temporoparietal regions. Because faces are differentiated from objects by around 120 milliseconds and structural encoding of faces begins between 140 and 200 milliseconds, the authors interpret these differences as reflecting atypical early facial encoding rather than later conscious processing.

Spectral decomposition of the EEG signal revealed a further difference in the theta band, oscillations between 4 and 8 hertz that occupy the same time window as the N170 component of face processing. Theta power differences between groups were greatest at 150 milliseconds after the window opened, and typically developed participants showed significantly higher theta power in the right temporoparietal electrodes P8 and CP6, both corrected at p equals 0.018. Theta rhythm is thought to orchestrate alternating phases of selective attention and exploratory sampling, with its trough aligned to focused fixation and its crest providing the window of highest probability for attentional shifting. Reduced theta power in the psychosis group may therefore relate to the atypical gaze behavior the eye tracker documented.

That gaze behavior itself differed between groups. Fifty-five participants contributed usable eye tracking data, and during the face gaze period the psychosis group showed significantly shorter fixation durations on the lab partner’s face and a significantly higher number of saccades per second. Spontaneous facial mimicry, the unconscious tendency to mirror a partner’s expressions, was also reduced: the group-level correlation between the first principal component of the participant’s facial action units and the lab partner’s was 0.12 for the psychosis dyads compared with 0.30 for the typically developed dyads, and a pooled variance t-test on individual correlation values confirmed the difference at p equals 0.035. Importantly, the lab partner’s own facial expressions did not differ significantly in magnitude between the two groups, confirming that the stimuli each group faced were equivalent.

The authors are careful to note the limitations of the work. The study is cross-sectional, so the link between atypical neural responses and behaviors such as gaze fixation, saccades and facial mimicry is strictly correlational. The groups differed in sex composition, with the psychosis group predominantly male and the control group predominantly female, though analyses treating sex as a covariate found no significant effects in the regions of interest. The smart glass opening introduces a non-social visual change alongside the face stimulus, and the absence of a non-social control condition such as a robot or mannequin is acknowledged as a major limitation, though the authors argue that the right temporoparietal clustering of differences is more consistent with face processing than with responses to sudden transparency changes, which would be expected to cluster in occipital and frontal channels. Subcortical nodes of the face processing system remain invisible to fNIRS because near infrared light penetrates only the superficial cortex. Still, the convergence of hemodynamic, electrophysiological and behavioral differences at the same right temporoparietal site suggests that ecologically valid, multimodal measurement can isolate neural hubs that single-modality or passive-viewing designs miss, and the authors propose that first episode psychosis, being a critical neuroplasticity window, could be an ideal stage for future causal probing studies with neuromodulation.

Subject of Research: Atypical neural and behavioral responses during live face-to-face interaction in first episode psychosis

Article Title: A multimodal approach to identify atypical face processing in first episode psychosis (early schizophrenia spectrum disorders)

Article References: Purushothaman, D., Noah, J. A., Zhang, X., Singh, R., Tek, C., Srihari, V. H., & Hirsch, J. (2026). A multimodal approach to identify atypical face processing in first episode psychosis (early schizophrenia spectrum disorders). Frontiers in Neuroimaging, 5, Article 1931240. https://doi.org/10.3389/fnimg.2026.1931240

Image Credits: AI Generated

DOI: 10.3389/fnimg.2026.1931240

Keywords: first episode psychosis, schizophrenia spectrum disorders, fNIRS, EEG, live face processing, supramarginal gyrus, theta oscillations, facial mimicry, eye tracking, PANSS negative symptoms, two-person neuroscience, social cognition