Face pareidolia is enhanced by 40 Hz transcranial alternating current stimulation (tACS) of the face perception network

Annalisa Palmisano; Giulio Chiarantoni; Francesco Bossi; Alessio Conti; Vitiana D’Elia; Serena Tagliente; Michael A Nitsche; Davide Rivolta

doi:10.1038/s41598-023-29124-8

. 2023 Feb 4;13:2035. doi: 10.1038/s41598-023-29124-8

Face pareidolia is enhanced by 40 Hz transcranial alternating current stimulation (tACS) of the face perception network

Annalisa Palmisano ^1,^✉,^#, Giulio Chiarantoni ^1,^#, Francesco Bossi ², Alessio Conti ¹, Vitiana D’Elia ¹, Serena Tagliente ¹, Michael A Nitsche ^3,⁴, Davide Rivolta ^1,⁵

PMCID: PMC9899232 PMID: 36739325

Abstract

Pareidolia refers to the perception of ambiguous sensory patterns as carrying a specific meaning. In its most common form, pareidolia involves human-like facial features, where random objects or patterns are illusionary recognized as faces. The current study investigated the neurophysiological correlates of face pareidolia via transcranial alternating current stimulation (tACS). tACS was delivered at gamma (40 Hz) frequency over critical nodes of the “face perception” network (i.e., right lateral occipito-temporal and left prefrontal cortex) of 75 healthy participants while completing four face perception tasks (‘Mooney test’ for faces, ‘Toast test’, ‘Noise pareidolia test’, ‘Pareidolia task’) and an object perception task (‘Mooney test’ for objects). In this single-blind, sham-controlled between-subjects study, participants received 35 min of either Sham, Online, (40Hz-tACS_ON), or Offline (40Hz-tACS_PRE) stimulation. Results showed that face pareidolia was causally enhanced by 40Hz-tACS_PRE in the Mooney test for faces in which, as compared to sham, participants more often misperceived scrambled stimuli as faces. In addition, as compared to sham, participants receiving 40Hz-tACS_PRE showed similar reaction times (RTs) when perceiving illusory faces and correctly recognizing noise stimuli in the Toast test, thus not exhibiting hesitancy in identifying faces where there were none. Also, 40Hz-tACS_ON induced slower rejections of face pareidolia responses in the Noise pareidolia test. The current study indicates that 40 Hz tACS can enhance pareidolic illusions in healthy individuals and, thus, that high frequency (i.e., gamma band) oscillations are critical in forming coherent and meaningful visual perception.

Subject terms: Human behaviour, Perception, Neurophysiology

Introduction

Pareidolia refers to the illusory perception of meaningful shapes from random or ambiguous stimuli. In the visual domain, individuals often report seeing images of faces, animals, or objects in random scenes, such as the Man in the Moon, the Moon Rabbit, and Jesus in a toast¹. From an evolutionary perspective, among the multitude of random stimuli we perceive in everyday life, others’ faces represent the most relevant stimuli for social relations, and our visual system seems to be particularly sensitive to facial configurations^2–4. Thus, individuals’ tendency to see human faces in clouds, mountains, or rocky discontinuities (i.e., face pareidolia) might be linked to our innate preference for face-like stimuli⁵, as well as to the existence of a face-detection system across primates^6,7. It should be noted that monkeys do not experience pareidolia as humans do, which could be due to a human-unique aptitude for anthropomorphizing objects with face-like patterns⁸. As in real face perception, individual differences characterize face pareidolia^9,10, with some people heavily perceiving face illusions while others reporting weaker effects^11–14.

Spatio-temporal features of the human face perception system have been extensively investigated with neuroimaging techniques such as functional Magnetic Resonance Imaging (fMRI) and electro/magneto-encephalography (EEG/MEG)^15–17. Face perception relies on a network of brain regions forming a “core” system (i.e., face-specific occipito-temporal areas)^18,19, and an “extended” system of prefrontal regions^20–22. Perception of real and illusory (i.e., pareidolic) faces show similar activity patterns²³, with “core” regions being both quickly (within 200 ms) engaged by pareidolic faces (i.e., illusory faces seen from random noise) and real faces^1,24–28. As seen in real face perception²⁹, and face mental imagery³⁰, face pareidolia requires the interaction between bottom-up and top-down paths (i.e., between visual areas and prefrontal regions)¹, to resolve perceptual decisions under uncertainty³¹, and give rise to conscious perception³². Furthermore, high-frequency (> 30 Hz) and low-amplitude neural oscillations in the gamma-band range (GBO)³³, as recorded with EEG/MEG^34–36, represent a potential neurophysiological correlate of face pareidolia. In fact, GBO mediate the general construction of coherent perceptual representations based on the integration of visual information^37–39.

Albeit the neuroimaging evidence reviewed above provides information about the localization and timing of face pareidolia, there is so far a lack of causal evidence of its neurophysiological underpinnings. Thus, the main aim of the current study is to test whether the exogenous entrainment of GBO within “core” and “extended” face regions via transcranial alternating current stimulation (tACS), by facilitating perceptual grouping and visual integration, can causally enhance pareidolic face illusions in a sample of healthy volunteers. This would be in line with neurophysiological evidence indicating that clinical populations experiencing frequent face-pareidolia, such as first-episode psychotic patients, are characterized by cortical hyperexcitability⁴⁰ and enhanced GBO⁴¹ in face-sensitive regions. tACS consists in delivering a weak sinusoidal electric current between two or more scalp electrodes⁴² that can be applied at biologically relevant frequencies. In line with the view that tACS selectively interacts with endogenous brain oscillations and related functions^43,44, frequency-tuned tACS can affect, for instance, visual perception^45,46, memory^47–49, problem-solving^50,51 and high-level visual cognition⁵².

The study of illusory face perception is rendered complex by the literature inconsistencies on the concept of face pareidolia and its operationalization. Indeed, it has been used to refer to faces seen in face-like patterns in the environment, as well as in meaningless noise patterns. Accordingly, previous studies investigated pareidolia with pure noise images (i.e., in which no face is present)¹, images with patterns resembling faces⁵³, pictures of environments with and without face-like patterns⁵⁴, and Mooney stimuli (i.e., black and white images from photographs of faces taken in a dark-contrasted environment vs. scrambled)^12,55. These latter have been consistently adopted in the literature, as the perception of Mooney stimuli is at the interplay between sensory processing, mental imagery, and visual working memory^56,57. In light of this heterogeneity, we adopted different tasks potentially catching pareidolia from multiple viewpoints. Both the frequency rate of illusory (i.e., non-existing) perceived faces⁵³, as well as changes in reaction times (RTs)⁵⁸, indicate pareidolia occurrence.

In line with previously reported tACS effects on both accuracy and RTs (e.g., visual detection tasks)^59,60, we hypothesized that tACS in the gamma (i.e., 40 Hz) frequency over critical face nodes of the “core” (e.g., electrode PO8 over the right occipito-temporal face-sensitive areas) and “extended” (e.g., electrode FP1 over the prefrontal cortex (PFC)) face network causally enhances face pareidolia in healthy participants. In other words, we expected 40 Hz tACS to cause more and/or faster “face answers” from scrambled/random-noise visual stimuli. Furthermore, since previous research has never directly investigated the effects of timing of tACS on visual cognition, we tested the behavioural effects of both online (i.e., tACS during task execution) and offline (i.e., tACS before task execution) neuromodulation (see⁶¹ for a similar approach).

Methods

Participants

A sample of 75 healthy volunteers (37 females; mean age 22.36 ± 2.42 SD) with normal or corrected-to-normal vision, and without any recorded history of psychiatric or neurological disorders were recruited for this single-blind, sham-controlled, between-subjects study. All participants were naïve to the research hypotheses and the experimental conditions. Participants were assigned to one of three groups receiving different stimulation protocols (see the section below). Prior to the testing session, they received a verbal and written explanation of the procedure and the potential adverse effects of brain stimulation (e.g., itching and tingling skin sensation, skin reddening, headache). Participants gave their written informed consent to participate. The study was conducted according to the ethical standards of the World Medical Association Declaration of Helsinki, and the study protocol received approval from the Ethics Committee of the University of Bari ‘Aldo Moro’ (protocol number: ET-19-01).

Experimental design

Participants were divided into three groups, receiving sham, or online (40Hz-tACS_ON) and offline (40Hz-tACS_PRE) 40 Hz tACS. Counterbalancing was applied within the Mooney tests, Noise pareidolia test, and Pareidolia task, with Mooney tests administered consecutively. Participants were assigned to one of three gender-matched groups of 25 participants each, receiving sham, 40Hz-tACS_ON, and 40Hz-tACS_PRE, respectively. All participants completed five tasks requiring circa 30 min of completion (see below for details on counterbalancing). Since females show similar cortical excitability as males only during the follicular phase of the menstrual cycle (when progesterone levels are low and estrogen levels are high), they were tested during this phase⁶². Tasks were set up with SuperLab 5.0 (Version 5.0.5, Cedrus Corporation, USA) and administered on a Fujitsu computer running Windows 10, with a 1920 × 1080 pixels 23-inches monitor. Participants were seated at approximately 60 cm eye distance from the screen.

After signing informed consent, participants were invited to complete five tasks: (i) the Mooney test for faces⁶³, (ii) the ‘Toast test’¹, (iii) the Noise pareidolia test⁵³, and (iv) the ‘Pareidolia task’. The (v) Mooney test for objects was also administered to monitor for potential category-specific effects (see below for tasks description and results). In order to align the experimental design with tasks’ length (i.e., the Toast test lasting circa 30 min, which represents the time taken to complete the remaining four tasks), grouping was designed as follows: 25 participants performed the Toast test during tACS (i.e., online), while Mooney tests (faces and objects), Noise pareidolia test, and Pareidolia task were completed offline (in a counterbalanced order, with the Mooney tests administered consecutively); whereas 25 participants completed the blocks in the opposite order, that is, online Mooney tests (faces and objects), Noise pareidolia test, and Pareidolia task (in a counterbalanced order, with Mooney tests administered consecutively), and offline Toast test. Participants receiving sham stimulation (N = 25) randomly performed either the Toast test online or offline, following the structure of the real tACS groups (Fig. 1). In each task, participants were asked to press the ‘M’ or ‘Z’ key on a computer keyboard to respond, respectively, whether they saw or not a face. The meaning of the response buttons was counterbalanced across subjects. At the end of the experimental session, all participants were debriefed and asked to fill in a ‘tACS adverse effects questionnaire’ about potential uncomfortable sensations experienced during or after the stimulation⁶⁴.

Schematic representation of the study design. Participants were divided into three groups, receiving sham, *online* (40Hz-tACS_ON) or *offline* (40Hz-tACS_PRE) 40 Hz tACS. Counterbalancing was applied within the Mooney tests, Noise pareidolia test, and Pareidolia task, with Mooney tests administered consecutively and counterbalanced between each other.

Mooney test for faces

The Mooney test for faces was created using black and white images from photographs of faces taken in a dark-contrasted environment⁵⁵ (Fig. 2). A total of 180 stimuli (size: 160 × 230 pixels) were selected from the Schwiedrzik Database⁶³. Participants were presented with three types of face stimuli: 45 upright (i.e., in canonic orientation), 45 inverted (i.e., upside-down), and 90 scrambled (meaningless images created from the face pictures). Randomized stimuli appeared for 350 ms, preceded by a 200 ms red fixation cross. Participants were asked to indicate, as fast as possible, if each stimulus contained a face or not. Correct answers were recorded if participants made their decision within 1200 ms from stimulus onset. A brief practice session with 10 trials was completed before the test. The tasks each took approximately 8 min to complete.

A higher rate of erroneous face detections versus correct scrambled recognitions (i.e., lower accuracy for scrambled stimuli), and/or longer RTs for no-face responses than face detections in participants receiving real stimulation than sham could reflect a tACS-driven pareidolia proneness in the Mooney test for faces.

Mooney test for objects

The Mooney test for objects was created ad-hoc using photo-warp.com on pictures of various object categories (e.g., buildings, furniture, etc.), and following a standard procedure (i.e., increase of chromatic saturation, decrease of sharpness to remove details, and maximization of stimulus contrast)^40,63, with a total of 180 stimuli (size: 160 × 230 pixels). The test structure was analogous to the Mooney test for faces (See Fig. 2 for sample stimuli). Since the Mooney test for objects was included as a control task to test the potential face-specificity of our tACS protocol, we expected no group differences in the rate of erroneous object detections between sham and verum stimulation.

Toast test

The face pareidolia section of the Toast test was taken from Liu and colleagues¹. This task includes training and a testing phase. The training was divided into three blocks of increasing difficulty; the first block included 20 easy-to-detect faces (well-defined hairless faces in the middle of visual noise) and 20 pure-noise images (white–gray spots on a black background); the second block comprised 20 hard-to-detect faces (little defined hairless faces in the middle of visual noise) and 20 pure-noise pictures, the third block contained 40 pure-noise images. Stimuli in each block were randomized for each participant. After the training phase, participants completed four testing blocks (120 stimuli per block). Although all stimuli in the testing session were pure-noise, participants were led to believe that 50% of them contained faces, with increasing levels of detection difficulty between sessions (Fig. 3). Both training and test stimuli were presented for 600 ms at the centre of the monitor, preceded by a 480 × 480 pixel checkerboard for 200 ms. At the beginning of each block, participants were instructed to decide, as fast as possible, if the stimulus contained a face or not. It took approximately 30 min to complete the task.

A higher rate of pareidolic responses versus correct noise detections, and/or shorted RTs for (always erroneous) pareidolic responses than noise detections in participants receiving real stimulation versus sham could reflect a tACS-driven pareidolia proneness in the Toast test.

Noise pareidolia test

An adapted version of the Noise pareidolia test⁵³ was adopted. The stimuli were panels containing black and white spots (Fig. 4). In 20% of the cases, the spots were located across stylized, decentralized faces. A total of 80 stimuli were used: 64 pure noise, 8 upright faces, and 8 inverted faces. Randomized stimuli appeared for 200 ms and were preceded by a checkerboard presented for 200 ms. Participants indicated if the stimulus contained a face or not. Participants had 1200 ms to answer. A brief practice session was completed before the test. The task took approximately 5 min to complete.

A higher rate of erroneous face detections versus correct noise recognitions (i.e., lower accuracy for noise stimuli), and/or longer RTs for no-face responses than face detections in participants receiving real stimulation than sham could reflect a tACS-driven pareidolia proneness in the Noise pareidolia test.

Pareidolia task

The Pareidolia task was composed of 100 coloured photos (50 face-like scenes and 50 scenes without any evident face-like pattern) (Fig. 4). A pilot study was conducted to validate the task. Eight volunteers were asked to decide if a face-like pattern was present or not in each picture (with no time limit). Since none of the images were classified in the wrong category, all stimuli were included in the task. Randomized stimuli appeared for 300 ms each. Participants indicated if they saw or not a face-like stimulus. They had to answer within 2000 ms after the end of the stimulus presentation when the screen was white. A brief practice session was completed before the actual test. The task took approximately 6 min to complete.

A higher rate of faces perceived in no-face images and/or shorter RTs for face- than no-face responses in participants receiving real stimulation than sham could reflect a tACS-driven pareidolia proneness in the Pareidolia task.

tACS

40 Hz tACS was delivered by a battery driven, constant current stimulator (BrainSTIM stimulator; E.M.S. s.r.l.) via a pair of surface sponge electrodes (25 cm²) soaked in saline solution (0.9% NaCl) and applied to the scalp at the target location. Electrodes delivered an alternating current of 2 mA (peak to peak; current density: 0.08 mA/cm²) for 35 min. We adopted a bilateral bipolar-non balanced montage⁶⁵. The sites of stimulation were identified using the International 10–20 EEG system, with one of the electrodes placed over PO8 (right occipito-temporal cortex–“core” face regions) and the other over FP1 (left PFC–“extended” face regions)^22,66. In the Sham group, the stimulator was turned on for only 20 s. to elicit a short-lasting skin sensation. Overall, tACS applications in this study complied with safety guidelines⁶⁷, and none of the participants reported major complaints or intolerable discomfort during or after tACS.

Statistical analyses

All analyses were conducted using R Studio (R Team, 2015). Mixed-effects generalized linear models (GLMMs) on a binomial distribution were adopted to test for tACS effects on performance, with a critical α-error = 0.05. Descriptive statistics, results for random effects and fixed effects from each task are reported in the Supplementary Materials.

The experimental group (“Group”, 3 levels: sham, 40Hz-tACS_ON, 40Hz-tACS_PRE) served as between-subject independent variable. Participants’ “Response” (face-present responses vs. noise detections) was the dependent variable of the Toast test. Responses’ Accuracy served as a dependent variable for the Mooney tests for faces and objects, Noise pareidolia test, and Pareidolia task. The type of stimuli was included as an independent variable in interaction with Group for the Mooney tests (“Stimuli”: upright, inverted, scrambled), Noise pareidolia test (upright, inverted, noise), and Pareidolia task (faces, landscapes). The assessment of an increase/reduction of pareidolic illusions was derived from the Group*Stimuli interaction, revealing how participants accurately and rapidly perceived the stimuli as faces or not. The Mooney tests for faces and objects were analyzed separately as the two sets of stimuli (i.e., faces and objects) are not taken from the same database, with the latter being not yet validated.

Participants’ RTs were also evaluated as a dependent variable by using mixed-effects linear models (LMMs), with the same independent variables. All models used random intercepts on participants. Trials with RTs beyond ± 2 SD from the participant’s mean were discarded as potential outliers. The significance of each effect was estimated using the Satterthwaite approximation for degrees of freedom in LMMs and performing likelihood ratio tests (LRTs) with corresponding null models in GLMMs. Post-hoc comparisons (using Tukey HSD p value correction) were performed to probe statistically significant interactions.

Uncomfortable sensation questionnaire

At the end of the experimental session, all participants were debriefed and asked to fill in a ‘tACS adverse effects questionnaire’ about potential uncomfortable sensations experienced during or after the stimulation protocol⁶⁴. The Uncomfortable Sensation Questionnaire comprised 8 items, and participants were asked to assess their sensations through a scale ranging from 0 (‘lack of sensations’) to 4 (‘strong sensations’). Student’s t-tests performed on scores from the ‘tACS adverse effects questionnaire’ indicated a significant difference between groups, with participants in the tACS groups reporting, overall, higher discomfort sensations than the sham group t(73) = − 4.821, p < 0.001.

Results

Mooney tests for faces

Accuracy data from the Mooney test (faces) showed a non-significant main effect of Group on Response (χ²(2) = 3.12, p = 0.210). The Group*Stimuli interaction on Response reached statistical significance (χ²(4) = 41.98, p < 0.001), with post-hoc contrasts based on Group showing that, for scrambled stimuli only, participants in the 40Hz-tACS_PRE group made more mistakes (mean = 0.526) than those in the sham (mean = 0.629) (z = 3.268, p < 0.001), and 40Hz-tACS_ON (mean = 0.615) (z = − 2.761, p = 0.016) groups, thus indicating that scrambled stimuli were more often erroneously perceived as real faces (Fig. 5). This demonstrates stronger face pareidolia (~ 10% increase of illusory face perceptions) after offline 40 Hz tACS. Post-hoc contrasts based on Stimuli showed significant differences in performance among all groups (all |z|s > 8; all ps < 0.001), with upright stimuli showing higher accuracy (sham = 0.890; 40Hz-tACS_ON = 0.890; 40Hz-tACS_PRE = 0.896; mean = 0.892) than inverted (sham = 0.734; 40Hz-tACS_ON = 0.745; 40Hz-tACS_PRE = 0.768; mean = 0.749), than scrambled stimuli (sham = 0.629; 40Hz-tACS_ON = 0.615; 40Hz-tACS_PRE = 0.526; mean = 0.59) (i.e., the well-known ‘Face Inversion Effect’ (FIE), an indirect index of holistic face processing^68,69).

Group*Response interaction for the Mooney test for faces. Post-hoc contrasts based on Group. Participants in the 40Hz-tACS_PRE group exhibited worse performance in terms of accuracy for scrambled stimuli, which were more often misperceived as faces than in the Sham and 40Hz-tACS_ON groups (i.e., induced offline pareidolia proneness).

RT data showed a main effect of Stimuli (F(2,8983.9) = 815.849, p < 0.001). The Group*Stimuli interaction was also statistically significant (F(4,8983.9) = 21.112, p < 0.001). Post-hoc tests for the factor Stimuli showed differences in performance in all groups, in line with the FIE, with upright stimuli resulting in faster RTs than the inverted and scrambled ones (all |z|s > 4.5; all ps < 0.001). Post-hoc contrasts based on the factor Group did not reveal any differences (all |z|s < 1.5; all ps > 0.299).