A Neural Signature of Similarity Assessment between a Face and One’s Own Face

A Neural Signature of Similarity Assessment between a Face and One’s Own Face | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A Neural Signature of Similarity Assessment between a Face and One’s Own Face Gennady Knyazev, Alexander Savostyanov, Andrey Bocharov, Alexander Saprigyn, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7570992/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Facial similarity to one’s own face indirectly affects a person’s social behavior. In this study, we aimed to develop, using multivariate pattern analysis, a brain signature for assessing facial similarity to one’s own face (SFSS). We hypothesized that it should have at least two aspects: one related to the assessment of similarity of faces, and the other specifically to the assessment of similarity to one’s own face. To distinguish between these aspects, we included two tasks that used the same set of morphed images, but in one case the task was to assess similarity to one’s own face and in the other to another person’s face. The SFSS showed excellent correlation with behavioral data and a large effect size in predicting the instantaneous similarity score for each participant. Mediation analysis showed that it acts as a mediator between the degree of similarity objectively present in an image and the subjective similarity assessment. Principal component analysis allowed to separate brain regions associated with the assessment of face similarity in general and with judging the similarity to one’s own face. It appeared that the early visual cortex was particularly involved in the former, while the fusiform gyrus and orbitofrontal cortex were involved in distinguishing targets. The results show the fruitfulness of considering brain signatures as representations of complex psychological processes, which can be analyzed by decomposing them into components to represent different aspects of relevant psychological states. biomarker multivariate pattern analysis neural signature physical self face similarity Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction Face similarity assessment is one of the basic social skills that greatly influences the social behavior of an individual. In particular, facial similarity to one’s own face is a factor that may indirectly influence the judgment of a person’s trustworthiness (Hansen et al 2020 ). Assessing facial similarity to one’s own face is akin to self-recognition, the brain underpinning of which have been the subject of many recent studies (Asakage and Nakano 2023 ; Devue and Brédart 2011 ; Hu et al. 2016 ; Janowska et al. 2021 ; Knyazev et al. 2024 ; Molnar-Szakacs and Uddin 2023 ; Morita et al. 2014 ; Ramon et al. 2015 ; Sugiura et al. 2015 ; Tramacere 2022 ; Żochowska et al. 2021 ). Nevertheless, assessing facial similarity to one’s own face and recognizing one’s own face are conceptually different tasks. Therefore, it can be expected that the neural signatures of these tasks may partially overlap and partially diverge. The present study aimed to explore using functional magnetic resonance imaging (fMRI) the brain signature of the assessment of facial similarity to one’s own face. The process of assessing the similarity of a face to one’s own face has two aspects. The first is the assessment of the similarity of faces regardless of the object with which the similarity is assessed. The second aspect is associated specifically with the assessment of the similarity of a face to one’s own face. To distinguish between these two aspects, we included two tasks in the experimental design. These tasks used the same set of morphed images, but in one case the participant’s task was to evaluate the similarity to his or her own face and in the other case to the face of another person. Since the pioneering work of Kanwisher et al. ( 1997 ), it is generally accepted that face processing in the human brain is mainly associated with face-selective brain areas (FSBA), such as the fusiform face area (FFA), the occipital face area (OFA), and the posterior part of the superior temporal sulcus (pSTS-FA). Haxby et al. ( 2000 ) proposed a highly influential neurocognitive model of face processing, according to which OFA implements early stages of face processing and directs its output to FFA, which represents unchanging aspects of the face, and to pSTS-FA, which represents changing aspects of the face. The information from this network is further processed in the extended face network. Later findings (reviewed in Duchaine and Yovel 2016) prompted a revision of this model. In particular, these findings indicate that multiple pathways directly connect early visual cortex (EVC) with multiple face-selective regions. An important feature of face processing in the brain is the dominance of the right hemisphere in face perception. Superior behavioral performance in the left visual hemifield (the so-called left visual field bias) and a preferential activation of right FSBA during face processing have been reported repeatedly (Åsberg Johnels et al. 2022 ; Harrison and Strother 2021 ; Hougaard et al. 2015 ; Levine et al. 1988 ). Given this information, one would expect that the assessment of facial similarity in general should be reflected primarily in right-sided changes in EVC and FSBA. When it comes to the second aspect, that is, the assessment of the similarity of a face specifically to one’s own face, it can be assumed to be akin to various types of self-referential processing, the neural basis of which has been the subject of numerous neuroimaging studies (e.g., Northoff et al. 2006 , 2011 ). It is often emphasized that there are at least two aspects of the self, the physical self and the psychological self (Gillihan and Farah 2005 ). Recognizing one’s own face is a paradigmatic example of processing the physical self. Meta-analyses of neuroimaging studies have shown that these two aspects of self-processing engage both distinct and shared brain regions, the latter including the dorsal anterior cingulate cortex (ACC) and the left inferior frontal gyrus (IFG), extending to the insula (Hu et al. 2016 ). An important feature of self-processing is that it is frequently associated with the activation of reward circuits (e.g., Hassall et al. 2016 ; Sui and Humphreys 2015 ). This is particularly true for self-face processing (Chakraborty and Chakrabarti 2018 ; Devue et al. 2009 ; Ota and Nakano 2021 ; Platek et al. 2009 ; Zhan et al. 2016 ). Thus, one might expect that assessing similarity specifically to one’s own face should engage brain regions associated with self-referential processing, such as the ACC, IFG, and reward circuits, such as the orbitofrontal cortex (OFC, see e.g., Diekhof et al. 2012 ; Howard and Kahnt 2021 ). One of the most important questions of such kind of research is the approach to data analysis. The mass univariate approach (Worsley et al. 1996 ), which has dominated the field in recent decades and has been used in the vast majority of published studies, has been criticized (e.g., Zhou et al. 2021 ) and has recently tended to be supplanted by various variants of multivariate pattern analysis (MVPA) (Haynes and Rees, 2006 ; Norman et al. 2006 ; Pereira et al. 2009 ). In the study of brain-mental state associations, MVPA methods yield much greater effect sizes than the traditional mass univariate methods and good test-retest reliability (Kragel et al. 2018 ). However, these methods also have their limitations and should be used with caution (Haufe et al. 2014 ; Jabakhanji et al. 2022 ; Ritchie et al. 2019 ; Weiskopf 2021 ; Vigotsky et al. 2024 ). In this study, we used both the traditional and MVPA methods to analyze fMRI data recorded during presentation of self-to-other morphed face images (e.g., Uddin et al. 2005 ), where the other was a familiar or unfamiliar person. To identify spatial patterns of brain activity associated with face similarity assessment, we used machine learning classification based on whole brain volume, selected regions of interest (ROIs), or local neighborhood as implemented in the searchlight approach. We also used the representational similarity analysis (RSA) (Kriegeskorte et al. 2008 ) to reveal where in the brain the similarity between spatial patterns of activation matched the similarity between individual ratings of facial similarity. The brain’s response to viewed images can be seen as a mediator between the proportion of the self that is actually present in the image and the assessment of its similarity to one’s own face. Indeed, the binary decision (self or not-self) has been shown to be strongly correlated with the actual similarity of the image to one’s own face and partially mediated by the brain’s graded response to the proportions of self that are present in the images (Knyazev et al. 2024 ). In this study, we aimed to test whether the signature response acts as a mediator between the degree of image deviation from one’s own face and the rating of its similarity to one’s own face. The most relevant issues in the study of brain signatures (or ‘decoders’) of specific mental states or psychological processes are the consistency and reliability of the identified spatial patterns and their relationship to the brain activity associated with these states. To answer the first question, we compared the topography of brain maps created using traditional mass univariate and different MVPA methods, and tested the reliability of signature expression across participants and on retesting. The traditional and MVPA methods are based on different ideologies. The former implicitly assume a functional organization of the brain, implying that compactly localized activations in specific ‘functional’ centers are sufficient to explain the brain’s involvement in stimulus processing (Worsley et al. 1996 ). The latter suggest that the content of a mental state is ‘encoded’ in a broadly spatially distributed pattern of neural activity, and the most accurate prediction of relevant states can only be achieved by accounting for this broad distribution (Zhou et al. 2021 ). To address this issue, we compared the predictive performance of decoders based on the whole brain volume with that of decoders based on specific functionally relevant ROIs. The second question is more difficult to answer. Some authors argue that although decoders based on machine learning can reliably predict the mental states on which they have been trained, their relevance to brain activity remains questionable (Haufe et al. 2014 ; Jabakhanji et al. 2022 ; Ritchie et al. 2019 ; Weiskopf 2021 ; Vigotsky et al. 2024 ). It is argued that machine learning classification methods can pick up information that is useful for improving prediction accuracy but is irrelevant to the task at hand (Haufe et al. 2014 ). Others argue that the machine learning used to create decoders is not needed at all, since the predictive performance of these decoders is no better than that of the mass-univariate encoders used to train them (Jabakhanji et al. 2022 ). To address these issues, we compared the predictive performance of obtained decoders with that of mass-univariate encoders, as proposed by Jabakhanji et al. ( 2022 ), and by comparing its weights with ‘activation patterns’, as proposed by Haufe et al. ( 2014 ). It is also argued that the information extracted by machine learning algorithms is highly redundant and more or less evenly distributed throughout the brain (Jabakhanji et al. 2022 ). To test this, we examined the effect of spatial smoothing on prediction performance and tested how the number of voxels affects the effect size of prediction-outcome association, and whether randomly selected voxels can demonstrate prediction performance comparable to whole-brain prediction. 2. Methods 2.1. Participants The sample used to develop the self-face similarity signature (SFSS) consisted of 54 right-handed participants (28 women, mean age 22 years, SD = 2.4). To compare the SFSS with the self-face recognition signature (SFRS) we used the data described in (Knyazev et al. 2024 ). This sample consisted of 44 right-handed participants (26 women, mean age 26.5 years, SD = 9). Exclusion criteria in both cases were serious medical conditions, history of epilepsy, substance abuse or dependence, and all contraindications to fMRI. The study complied with the Declaration of Helsinki of the World Medical Association and was approved by the Ethics Committee of the Institute of Neurosciences and Medicine. All participants gave written informed consent to participate in the experiment. 2.2. Face similarity judgement experimental procedure Each participant was invited into the laboratory with a good acquaintance or friend of the same gender. In addition, another person of the same age and gender, unfamiliar to the participant, was invited. All participants were photographed. The photographs were taken in frontal projection and then a mask was applied to crop out all hair details. Before morphing, point distinctive features such as birthmarks or blemishes were removed from photographs. Morphing software ( http://alyssaq.github.io/face_morpher/ ) was used to create intermediate images for all pairwise combinations (i.e., me vs friend, me vs stranger, and friend vs stranger). Images were ranked on a scale from 0 to 5, where 5 is the target face (i.e. the face to which similarity is being assessed) and 0 is the face of another person without morphing. The intermediate gradations are the different stages of morphing. Participants were given the following instructions: “In this experiment, you will be shown pictures of faces and asked to rate the degree of similarity of each face to the target face.” The experiment consisted of two sessions. In the first session, the target was the participant’s own face, and in the second session - the face of his\her friend. The order of sessions was chosen at random for each participant. Each session consisted of three runs separated by two-minute rest intervals. Each run involved the presentation of two types of morphed images, i.e., morphed toward familiar (i.e., participant’s own or his\her friend’s face) or unfamiliar (taken from the database) face. The images were presented in random order. First, a cross appeared in the center of the screen, then after 1 ± 0.1 s - an image of a face, which remained on the screen for 2.5 s, after which the scale appeared, which allowed to select the degree of similarity. The scale ranged from 0 (not the target) to 10 (exactly the target). In data analysis, participants’ ratings were averaged for each of the six morphing stages. Five s were given for the selection stage. Only 2.5 s of face presentation were used for analyses. The delay between the start of the upcoming fMRI frame and the start of the trial varied randomly between intervals of 100 and 2350 ms. The self-face recognition experiment described in (Knyazev et al. 2024 ) used the same stimuli, but the participant’s task was to make a binary decision (i.e., me or not me). 2.3. fMRI Data acquisition and preprocessing fMRI data were acquired using an EPI sequence on a Philips Ingenia 7FN8GDI 3.0 T scanner (Amsterdam, Netherlands). The first five scans at the beginning of each run were discarded to ensure the effect of scanner equilibration (TR = 2.5 s, TE = 35 ms, flip angle = 90°, FOV = 100, 96×94 matrix, 25 slices of 5 mm thickness, no gap). T1-weighted high-resolution (1 mm) structural scans were obtained using a 3D MP-GR sequence (TR = 7.8 ms, TE = 3.76 ms, matrix 252×227). Preprocessing was performed using the SPM-12 toolbox and included slice-time correction, realignment using rigid body transformation, coregistration, normalization to the Montreal Neurological Institute (MNI) template, and resampling to 2×2×2 mm resolution. Because MVPA methods are sensitive to signal-to-noise ratio (Dimsdale-Zucker and Ranganath 2018 ), we additionally denoised the data using the GLMsingle software that enables accurate estimation of single-trial fMRI responses (Prince et al. 2022 ). A GLM design matrix with separate regressors for each trial presentation and six realignment parameters as additional regressors of no interest was constructed, and for each voxel, a custom hemodynamic response function was identified from a library of candidate functions. Cross-validation was used to derive a set of noise regressors from voxels unrelated to the experiment, and finally the betas were regularized on a voxel-wise basis using ridge regression (Prince et al. 2022 ) and smoothed (full-width half-maximum 4 mm). 2.4. GLM analysis Univariate GLM analysis was conducted using the SPM-12 toolbox. Denoised beta weights were averaged across trials and runs for each participant separately, further smoothed (full-width half-maximum 8 mm), and subjected to a second-level random-effects analysis to allow for population inferences. Three-way ANOVA was used to reveal the effects of three experimental factors, including the target (participant’s own face or his\her friend’s face), the familiarity of the opposite face (i.e., morphed toward familiar or unfamiliar person), and similarity ratings (six levels), and their interactions. 2.5. Deriving the signature To create the signatures of self-face similarity (SFSS) and friend-face similarity (FFSS), we used the Least Absolute Shrinkage and Selection Operator-Regularized Principal Components Regression (LASSO-PCR; Wager et al. 2011 ), as implemented in CanlabCore Tools v2 ( https://github.com/canlab/CanlabCore , accessed on 30 December 2024). This is a regression-based method, which is frequently used for predicting continuous variables, such as e.g. self-reported ratings of pain intensity (e.g., Wager et al. 2013 ). To derive the SFSS and FFSS across contexts, we aggregated the experimental conditions with familiar and unfamiliar opposite faces. In additional analyses the signatures were generated for each condition separately. In each participant, in each run, and for each target (i.e., me and friend) separately, the beta images obtained in single trials were averaged across the six morphing stages and used as input to the LASSO-PCR algorithm to predict participants’ ratings, which also were averaged across the six morphing stages. Additional analyses were conducted in which participants’ ratings were averaged into five categories regardless of morphing stages, and the corresponding beta weights were averaged across these categories. Some participants did not use the full range of allowed ratings and these participants were excluded from these analyses. This left 40 participants in the ‘friend’ condition and 42 participants in the ‘me’ condition. Comparison of the results did not reveal a significant difference, so we report the results obtained on the full dataset. We employed a 5-fold cross-validation, in which all participants were randomly assigned to 5 subsamples of 10 or 11 participants and the model is fitted on 4 of the subsamples and tested on the left-out subsample. This is repeated 5 times with each subsample being the testing set once to assess cross-validation performance (Wager et al. 2011 ). Jabakhanji et al. ( 2022 ) argue that machine-learning-derived fixed-weights decoders are not required to successfully predict a behavioral outcome, which can be achieved based on the information present in the initial beta weight map used to train the decoders (i.e., encoders). To test this, we compared predictive performance of SFSS with that of the initial beta-map and the contrast-based (i.e., me vs not me) beta-map. 2.6. Searchlight regression analysis Searchlight analysis was performed using CoSMo MVPA toolbox ( http://www.cosmomvpa.org/ ) (Oosterhof et al. 2016 ). For each voxel in the brain, a spherical neighborhood (SN) comprising the 200 voxels closest to the central voxel in the brain grey matter mask was defined and the LASSO-PCR algorithm described above and a 5-fold cross-validation scheme were used to predict participants’ scores based on each SN. Prediction accuracy for each participant was defined as the correlation between that participant’s ratings and the responses calculated by the LASSO-PCR algorithm for that participant. The significance of these correlations was tested using a one-tailed permutation test (P < 0.001, 20,000 permutations) (Crosse et al. 2024 ). For SNs that passed the test, the overall prediction-outcome correlation was summarized at its center voxel. 2.7. The whole brain vs. ROI analysis The above-described procedure was performed using as input the beta images covering either the whole brain (limited to the gray-matter mask without the brain stem and cerebellum) or selected ROIs. We selected six large ROIs, each of which included a set of brain regions involved in functions related to face recognition, self-processing, or other putatively relevant processes. These regions were identified based on the fMRI meta-analytic resource Neurosynth ( https://identifiers.org/neurovault.collection:2099 , Yarkoni et al. 2011 ) and the search terms ‘early visual’ (136 studies, 949 voxels), ‘inferior occipital’ (116 studies, 690 voxels), ‘fusiform face’ (143 studies, 1920 voxels), ‘face recognition’ (79 studies, 1095 voxels), ‘self-referential’ (166 studies, 1075 voxels), and ‘medial orbitofrontal’ (121 studies, 1112 voxels). Note that ‘early visual’ ROI represented EVC, ‘inferior occipital’ ROI represented OFA, and ‘fusiform face’ ROI represented FFA. 2.8. Representational similarity analysis We used the CoSMoMVPA toolbox to perform RSA for the five selected ROIs (Oosterhof et al. 2016 ). The goal of RSA was to compare the similarity of patterns of blood oxygen level dependent (BOLD) voxel-wise responses to each picture with the similarity of participants’ ratings of these pictures. To construct neural representational dissimilarity matrix (RDM), we used the inverse of Spearman rank correlation coefficients (Popal et al. 2019 ). For each subject separately, multi-voxel activation patterns from a ROI were averaged for each rating category (six categories) across runs and conditions (i.t., familiar and unfamiliar opposite face) for the ‘self’ and ‘friend’ targets separately and used to calculate inverse correlations between these averages. Behavioral RDMs were calculated as the absolute difference between participant’s similarity scores averaged respectively. These behavioral RDMs were used as the reference RDM, and the relatedness of each ROI RDM to the reference RDM was calculated for each subject using Spearman rank correlations between the upper triangles of the respective RDMs (Nili et al. 2014 ). In all cases the resulting subject-specific Fischer-z-transformed correlation coefficients were entered into a one-sample permutation t-test (20,000 iterations) against the null hypothesis of a mean of zero (Crosse et al. 2024 ). 2.9. Mediation analysis According to the formal mediation model outlined in Baron and Kenny ( 1986 ), we need to have significant relationships between the independent variable (here, the morphing stages), the dependent variable (here, the facial similarity rating), and the mediator (here, the signature response). In addition, the relationship between the independent and dependent variables should decrease or become not significant after controlling for the mediator effect. To test the mediation model, we used the mediation function from the M3 Mediation Toolbox ( https://github.com/canlab/MediationToolbox , Wager et al. 2009 ). For this analysis, we combined the familiar and unfamiliar opposite face conditions, thus obtaining 108 trials for each participant. The mediation path model was fitted in each participant and, at the second level, the significance of effects in the group of participants was tested using bias-corrected accelerated bootstrap tests (10,000 samples) (Shrout and Bolger 2002 ). 2.10. Statistical testing In the traditional GLM analysis, the main effect of each factor, as well as their interactions were estimated at a voxel-level threshold of p < 0.001 uncorrected for multiple comparisons and a cluster-level threshold of p < 0.05, FWE-corrected. After deriving the signature, for display only, we thresholded the signature map using a 10000-sample bootstrap procedure at q < 0.05, FDR corrected (Zhou et al. 2021 ). Statistical maps were visualized and cluster information (including location of activation peaks, the size of each cluster, and the closest Talairach label for each peak coordinate) was extracted using NeuroElf software ( http://neuroelf.net ). Haufe et al. ( 2014 ) using simulated data showed that high weight ascribed by a classification algorithm to a particular brain region does not necessarily mean that this region is directly involved in the cognitive process of interest. Therefore, a more accurate interpretation of the predictive brain regions could be achieved by comparing the multivariate patterns obtained by means of machine learning methods with ‘activation patterns’ showing the direction of the relationship between each brain region activity and the cognitive process of interest without controlling for other variables. In statistical terms, these activation patterns are similar to the so-called structure coefficients, whose usefulness in interpreting multiple regression results has been recognized (Courville and Thompson 2001 ). According to Haufe et al. ( 2014 ), voxels showing significant predictive weights and structure coefficients are important because they are both directly correlated with outcomes and are predictive after accounting for other brain regions. For computing within-participant activation patterns we used the formula presented in Zhou et al. ( 2021 ), implemented in CanlabTools’ fast_haufe function. The significant brain regions were determined using a one-sample t test thresholded at q < 0.05, FDR corrected. To test the predictive performance of signatures developed based on whole brain volume or selected ROIs, signature responses were calculated as dot product between the signature weights and the brain image of estimated activity (i.e., beta weights, Han et al. 2022 ). These responses were then correlated with behavioral data (e.g., similarity scores). In addition, the predictive performance of signatures was evaluated by prediction-outcome correlations based on the data obtained in the process of cross-validation. The significance of within-person correlations was tested using a one-sample permutation test (P < 0.001, 20,000 permutations) (Crosse et al. 2024 ), which also was used for statistical comparison of Fischer-z-transformed correlation coefficients. Short-term test-retest reliability of the mean signature response was determined by the intra-class correlation coefficient (ICC; Koo and Li 2016 ; Han et al. 2022 ). Reliability was assessed for all participants in three runs within one experiment. We used a two-way mixed-effects model with time as a fixed effect and participants as a random effect, as implemented in SPSS package. 3. Results 3.1. Behavioral data A repeated-measures ANOVA with the factors of target (me vs friend), familiarity (familiar vs unfamiliar opposite face) and similarity (six levels), and similarity scores as dependent variable showed significant main effects of familiarity (F 1,53 = 29.88, p < 0.001, η p 2 = 0.361) and similarity (F 5,265 = 1033.01, p < 0.001, η p 2 = 0.951), as well as an interaction between similarity and familiarity (F 5,265 = 19.83, p < 0.001, η p 2 = 0.272). As shown in Fig. 1 , for both targets, similarity scores are lower when an unfamiliar opposite face is used for morphing. Similarity scores strongly correlated with morphing stages (mean r(54) = 0.94, SD = 0.04). Paired samples permutation t-tests on Fischer-z-transformed correlation coefficients found no difference between different combinations of target and familiarity conditions (all ps > 0.12). 3.2. Univariate GLM analysis Three-way ANOVA was performed on denoised beta weights with the factors of target (‘seld’ vs ‘friend’), familiarity (familiar vs unfamiliar opposite face) and similarity (six levels). The main effect of target did not produce significant effects. The main effect of familiarity was significant in the left middle temporal gyrus (T 1,1272 = 4.75; x = -45, y = -61, z = 17; K E = 218; P FWE−corr = 0.001). The main effect of similarity was significant in a cluster spanning many brain regions and centered in the right lingual gyrus (F 5,1272 = 196.39; x = 12, y = -85, z = -7; K E = 20147; P FWE−corr < 0.001). Other effects were not significant. Figure 2 shows the localization of the main effect of similarity, as well as contrast estimates and 90% confidence intervals of the left and right lingual gyrus response at each similarity rating point. It can be seen that the right lingual gyrus response shows a gradual increase in activation with increasing similarity ratings, while the left lingual gyrus response shows the opposite dynamics. 3.3. Deriving the self-face (SFSS) and friend-face (FFSS) similarity signatures Figures 3 A and 3 B show SFSS and FFSS thresholded at q < 0.05, FDR corrected. Overall prediction-outcome correlation based on the data obtained in the process of cross-validation was r(1944) = 0.74 for SFSS and r(1944) = 0.72 for FFSS. Mean within-person correlation was r(36) = 0.73 ± 0.15, p < 0.001 for SFSS and r(36) = 0.72 ± 0.11, p < 0.001 for FFSS. In both cases, signature responses showed a perfect (r = 1) correlation with similarity scores obtained in the same condition. SFSS responses also significantly correlated with similarity scores in the ‘friend’ condition (r(1944) = 0.66, t = 32.5, p < 0.001), and FFSS responses significantly correlated with similarity scores in the ‘self’ condition (r(1944) = 0.54, t = 19.6, p < 0.001). A comparison of Figs. 3 A and 3 B shows that the distributions of most predictive weights look similar, however, the overall correlation between SFSS and FFSS patterns was only r(47158) = 0.175, p < 0.001. Both patterns showed comparable correlations with the distribution of F values obtained in the univariate GLM analysis of the main effect of similarity and presented in Fig. 2 A – r(47158) = 0.127, p < 0.001 and r(47158) = 0.124, p < 0.001 for SFSS and FFSS, respectively. Figures 3 C and 3 D show the ‘activation patterns’ (i.e., structure coefficients) averaged across participants and FDR-thresholded at q < 0.05. It appears that most regions with significant model weights also show significant activation. Spatial correlations between signatures and activation patterns were r(47158) = 0.107, p < 0.001 and r(47158) = 0.355, p < 0.001 for SFSS and FFSS, respectively. The overlap between the two signatures and corresponding activation patterns is shown in Figs. 3 E and 3 F. In both cases most prominent clusters included the right and left lingual gyrus. 3.4. Exploring the properties of SFSS 3.4.1. Predictive performance of encoders Jabakhanji et al. ( 2022 ) argue that multivariate fixed weights decoders perform no better than the GLM-derived encoders. To test this, we assessed the predictive performance of the GLM-derived map of the main effect of similarity. When used as a decoder, this map’s expression showed a correlation of 0.22 (p < 0.001) with similarity scores in the ‘self’ condition. For within-person prediction, mean correlation was 0.25 (t(54) = 8.5, p < 0.001). 3.4.2. Spatial smoothing of voxel weights To investigate the effect of spatial smoothing of the SFSS weights on its performance, we used a Gaussian filter with increasing width, from 1 mm to 20 mm. At 1 mm smoothing, the performance deterioration was very small (r(1944) = 0.9948), but the difference was already significant, as shown by a permutation paired samples t-test for all participants (t 54 = 9.12, p < 0.001). Performance deteriorated further with increasing filter width, reaching a prediction-outcome correlation of 0.71 at 20 mm smoothing. We also constructed a ‘sign’ version of SFSS in which all positive voxels were assigned a value of + 1 and negative voxels were assigned a value of -1. The performance of this version was significantly worse than the original SFSS (r = 0.57). 3.4.3. Effect of the number of features/voxels Next, we analyzed the effect of the number of voxels randomly selected in the whole brain on prediction success. An increasing number of voxels (starting at 100 and in increments of 100) were repeatedly (100 iterations for each step) randomly selected from the whole brain and used to predict the similarity scores in the ‘self’ condition. As can be seen in Fig. 4 , the model performance is unstable below 2500 voxels, then increases asymptotically and reaches a plateau, approaching (but not reaching) the performance of a signature based on whole brain analysis when a random sample of about 5000 voxels is selected. Then, following the suggestion of Zhou et al. ( 2021 ), 10,000 voxels were randomly selected in the brain. The difference between averaged over 1000 iterations performance of 10,000 randomly selected voxels and all voxels in the brain was still significant across participants, as shown by the paired samples permutation test (t 54 = 4.2; p = 0.0001). Further increase of the number of voxels showed that the difference remained to be significant up to 25,000 voxels (p = 0.020), but was not significant with 30,000 voxels (p = 0.477). 3.5. Within- and between-person SFSS reliability The overall short-term (for the three runs within one experiment) test-retest reliability of SFSS response averaged across trials was excellent (ICC = 0.97, F 53,106 = 33.6, p < 0.001). For run 1 versus run 3 it also was excellent (ICC = 0.94, F 53,53 = 17.4, p < 0.001). An important question is to what extent a population-level signature such as the SFSS can predict each participant’s similarity score on each trial. To test this, we applied the SFSS to each participant’s single-trial beta map to obtain instantaneous signature responses in each trial. For each participant separately, Pearson’s correlation coefficient was calculated between these responses and the behavioral responses on each trial. These coefficients were Fischer-z-transformed and entered into a one-sample t-test to assess its statistical significance. Mean (SD) correlations were 0.78 (0.06), t 53 = 93.0, p < 0.001, Cohen’s d = 13.0. In all participants, correlations were significant at q < 0.05, FDR-corrected level. We also tested the performance of SFSS using pooled single-trial data for all participants. In this case, the correlation was 0.79 (0.77–0.80), Cohen’s d = 2.58. Another question is how well the SFSS can predict individual differences in the ability to assess facial similarity. To answer this question, we assessed each participant’s sensitivity to the similarity of the images to their own face as a correlation between the morphing stages and similarity ratings. The same indices were calculated for SFSS responses. The Pearson correlation coefficient for behavioral and neural indices of sensitivity was r(54) = 0.33, p = 0.017, Cohen’s d = 0.70. 3.6. Mediation analysis The results reported so far show that facial similarity ratings correlate with the degree of image morphing. SFSS response also correlated with the degree of image morphing r(1944) = 0.88. In addition, SFSS response correlates with similarity ratings, as shown by the prediction-outcome correlation described above. Thus, the initial conditions for testing of the mediation model are fulfilled. The mediation analysis showed a significant effect of mediation (B = 2.01, STE = 0.06, t 53 = 36.26; p = 0.0004). The strength of association between the morphing stage and the facial similarity ratings (B = 2.01, STE = 0.06, t 53 = 36.32; p = 0.0004) became no significant (B = 0.00, STE = 0.00, t 53 = 0.56; p = 0.579) when controlling for the effect of the signature response. Figure 5 shows results of bootstrap tests of the mediation model path coefficients. It can be seen that the significant slope c (degree of morphing→similarity scores without controlling for signature response) becomes no significant ( c’ ) while controlling for signature response. 3.7. Searchlight analysis Figure 6 shows the results of the searchlight regression analysis. The highest accuracy scores were found in the right and left lingual gyrus. 3.8. ROI-based signatures Table 1 shows prediction-outcome correlations obtained for each ROI in the process of cross-validation in the ‘self’ and ‘friend’ conditions. The highest correlations were observed for the EVC and OFA, while the FFA showed the smallest correlation. The correlation between the size of the ROI and its predictive performance was not significant (p > 0.4). Table 1 Prediction-outcome correlations for ROI-based signatures in the ‘self’ and ‘friend’ conditions. ROI ‘self’ ‘friend’ Early visual 0.69 0.66 Inferior occipital 0.34 0.32 Fusiform face area 0.04 0.05 Face recognition 0.18 0.16 Self-referential 0.22 0.20 Medial orbitofrontal 0.06 0.09 All correlations are significant at p < 0.05, FDR-corrected. 3.9. Representational similarity analysis Spearman correlations between ROI RDMs and behavioral RDMs averaged across participants in the ‘self’ and ‘friend’ conditions are presented in Table 2 . Significance of correlations was tested using a one-sample permutation t-test across participants. The highest correlation in both conditions could be noted for EVC. For self-referential and MOC ROIs, correlations are smaller in the ‘friend’ than in the ‘self’ condition. For MOC the correlation was significant in the ‘self’, but was not significant in the ‘friend’ condition. For self-referential ROI it was significant in both cases, but a paired-samples permutation t-test showed that it was significantly higher in the first case (t 52 = 2.01, p = 0.023). Table 2 RSA results. Spearman correlations between ROI RDMs and behavioral RDMs averaged across participants. Significance of correlations was tested using a one-sample permutation t-test across participants. ROI ‘self’ ‘friend’ Early visual 0.64* 0.62* Inferior occipital 0.22* 0.23* Fusiform face area 0.21* 0.20* Face recognition 0.14* 0.12* Self-referential 0.25* 0.13* Medial orbitofrontal 0.08* 0.001 * correlation is significant at p < 0.05, FDR-corrected. 3.10. Decomposing SFSS and FFSS into common and distinct components As we discussed in the Introduction, the mental processes whose signatures are SFSS and FFSS must have both common and distinct components - the former are related to the evaluation of similarity of faces regardless of the target with which the similarity is evaluated, and the latter related specifically to the target of assessment. To decompose the SFSS and FFSS into these components, we used principal component analysis (PCA). The first component, which presumably accounted for the common features of SFSS and FFSS, explained 59% of the variance. Using the pcares Matlab function we created signatures of both reconstructed (REC) and residual (RES) patterns after extracting this component. Thresholded maps of these patterns are shown in Figs. 7 A and 7 B. It can be seen that the REC map has strong positive values in the right lingual gyrus and strong negative values in the left lingual gyrus. The RES pattern has most prominent positive and negative values in occipital, temporal, and orbitofrontal regions. Responses of REC and RES patterns were then calculated in the ‘self’ and ‘friend’ conditions and correlated with similarity scores. Correlations between REC responses and similarity scores were 0.88 and 0.92 in the ‘self’ and ‘friend’ conditions, respectively. For RES they were − 0.52 and 0.55, respectively. This confirms that REC reflects features common to both conditions, while RES reflects features that distinguish them. Next, we regressed out REC from beta weights in original data and trained the LASSO-PCR algorithm with 5-fold cross-validation to predict similarity scores in the ‘self’ and ‘friend’ conditions. The obtained purified self-face similarity (PSFSS) signature looked similar to RES. Correlations between PSFSS responses and similarity scores were 1.0 and − 0.09 in the ‘self’ and ‘friend’ conditions, respectively. Correlations between purified friend-face similarity signature (PFFSS) responses and similarity scores were − 0.07 and 0.99 in the ‘self’ and ‘friend’ conditions, respectively. Thus, both PSFSS and PFFSS predict similarity scores in their own conditions no worse than the original SFSS and FFSS, but do not predict similarity scores in the opposite condition. PSFSS and PFFSS patterns correlated negatively with each other (r(47158) = -0.76). We then analyzed the overlap between the positive and negative weights of PSFSS and PFFSS. They were transformed into z-scores and thresholded at z = 1.96. Next, overlaps were analyzed for positive vs positive, negative vs negative, and positive vs negative weights separately. There were no overlaps for positive vs positive and negative vs negative weights. Only overlaps for positive vs negative weights were found in the fusiform and orbital gyri (Fig. 7 C). 3.10. Overlap with the self-face-recognition signature (SFRS) The SFRS was developed previously using the same analytic approach and experimental paradigm, which was similar to the paradigm of the present study in all aspects except one: on each trial, participants were required to make a forced choice decision (‘me’ or ‘not me’). We first tested whether the SFSS response could predict self-recognition in a forced choice task, and conversely, whether the SFRS response could predict a face’s similarity to one’s own face score in this study task. A one-sample permutation t-test across participants was used to determine the significance of the correlation between prediction and outcome in each participant. It showed a significant prediction-outcome correlation in both cases (mean r = 0.09, t 53 = 2.11, p = 0.041 and mean r = 0.17, t 53 = 5.98, p < 0.001 for SFSS and SFRS, respectively). When FFSS and, respectively, friend’s face similarity data were used instead of SFSS and self-face similarity data, the prediction-outcome correlations were not significant (mean r = 0.006, t 53 = 0.16, p = 0.874 and mean r = 0.04, t 53 = 1.57, p = 0.12 for FFSS and SFRS, respectively). To identify commonalities in SFSS and SFRS, we again used PCA. The first component, which accounted for the common features of SFSS and SFRS, explained 98% of the variance. Using the pcares Matlab function we reconstructed the signature of this component. This signature response showed significant overall correlations with both self-recognition in a forced choice task, r(704) = 0.82, p < 0.001, and self-face similarity score in this study task, r(1944) = 0.71, p < 0.001. A thresholded map of this signature is shown in Fig. 7 D. It has positive and negative weights in the right fusiform gyrus and right and left orbitofrontal regions. It also showed positive weights in the right lingual gyrus and negative weights in the left lingual gyrus. 4. Discussion In this study, we aimed to explore the brain signature of the assessment of facial similarity to one’s own face (SFSS). We aimed to compare the results of a traditional mass-univariate data analysis approach with those of MVPA and to examine the properties of the signature obtained using MVPA methods. Besides, we hypothesized that this signature may have two components: one related to the assessment of face similarity independently of the object with which the similarity is assessed, and the other related specifically to the assessment of face similarity to one’s own face. To distinguish between these two components, we included a task in which participants rated the similarity of morphed images to their friend’s face. In addition, we compared the SFSS to a self-recognition signature previously derived in an experimental paradigm that does not require explicit face similarity assessment. We expected that face similarity assessment in general should be reflected predominantly in right-sided changes in EVC and FSBA, whereas similarity assessment specifically to one’s own face should engage brain regions associated with self-referential processing. Behavioral results show that participants were able to reliably judge the actual degree of similarity of morphed images to target faces independently of target and familiarity of the opposite face, although similarity scores tended to be higher when a familiar face was used. The familiarity factor also had a significant effect in univariate GLM analysis, which was observed in the left middle temporal gyrus, whose role in assessing face familiarity, presumably related to providing access to face-related semantic knowledge, has been noted previously (Elfgren et al. 2006 ; Pourtois et al. 2005 ). In general, however, GLM analysis results indicate that the observed effects are preferentially associated with the assessment of face similarity regardless of the target with which the similarity is assessed. This is evident from the fact that the main effect of target was not significant, whereas the main effect of similarity spanned multiple brain regions, reaching a maximum in the right EVC. The right-sided dominance of the similarity effect is consistent with the known dominance of the right hemisphere in the perception of faces (Åsberg Johnels et al. 2022 ; Harrison and Strother 2021 ; Hougaard et al. 2015 ; Levine et al. 1988 ). Notably, right-sided EVC activation gradually increases with increasing similarity ratings, while left-sided EVC activation exhibits the opposite dynamics (see Fig. 2 B). This shows that the initial information for face similarity assessment is already present in EVC (see e.g., Lyons and Morikawa 2020 ; Yue et al. 2012 ), which then passes to higher levels of face similarity processing (Duchaine and Yovel 2016). Comparison of the results of univariate and multivariate analyses shows that the prognostically most significant SFSS weights coincide with the major loci identified in the GLM analysis. This mostly relates to the right and left EVC areas. Notably, the right-sided EVC contains positive SFSS weights and the left-sided EVC contains negative weights, which is consistent with the dynamics identified in the GLM analysis. Comparing the SFSS weights to the thresholded ‘activation pattern’ again shows that the right and left EVC regions are coincidence locations. These regions also showed the highest accuracy rates in the local neighborhood-based searchlight regression analysis. The RSA results and the prediction of similarity scores based on ROIs further support the idea that EVC areas are crucial for face similarity estimation. Indeed, the EVC provided an overall prediction-outcome correlation similar to that of SFSS (see Table 1 ), and showed the highest relatedness estimates in RSA (see Table 2 ). Noteworthy, FFA scarcely allowed the prediction of similarity scores and showed considerably lower relatedness estimates in RSA. Thus, the main findings from different analysis methods are consistent in showing the primacy of EVC for face similarity assessment. The overall prediction-outcome correlation obtained for SFSS is comparable to that obtained in other relevant studies (e.g., Knyazev et al. 2024 ; Zhou et al. 2021 ), and SFSS response showed a perfect correlation with behavioral data and excellent short-term test-retest reliability. Moreover, it showed a large effect size in predicting the instantaneous (i.e., in each trial) similarity score for each participant and was significantly associated with individual differences in sensitivity to the similarity of the images to participant’s own face. Most impressively, the mediation analysis showed that the SFSS response acts as a mediator between the degree of similarity objectively present in the image and the subjective assessment of similarity. All this suggests that the SFSS is a robust brain signature of the mental processes involved in assessing the similarity of face images to one’s own face. Following Jabakhanji et al.’s ( 2022 ) suggestion, we explored the properties of this signature. Our results contradict the claim that predictive performance of machine-learning-based decoders is always no better than that of the mass-univariate encoders used for their derivation. In our case, the predictive performance of the GLM-derived map of the main effect of similarity was significantly worse than that of SFSS. Spatial smoothing of the SFSS weights degraded its predictive performance as early as 1 mm smoothing, and the ‘sign’ version of the SFSS performed significantly worse than the original SFSS. Analyzing the effect of the number of randomly selected voxels showed that more than a half of voxels were needed to achieve the prediction success rate obtained using the whole brain. These properties of SFSS differ from those of SFRS described earlier (Knyazev et al. 2024 ). It can be hypothesized that they depend on the characteristics of the mental processes that are being decoded. In particular, self-face similarity assessment is a more difficult task than self-face recognition, and, respectively, the information needed to decode it must be more specific and structured. After confirming that the SFSS is a reliable brain signature for assessing similarity between a face and one’s own face, a legitimate question arises as to what mental processes it reflects. As we discussed in the Introduction, the mental processes involved in judging the similarity of images to one’s own face must have at least two components – one related to judging the similarity of faces independently of the object with which the similarity is judged, and one related specifically to judging the similarity of images to one’s own face. To separate these processes, we compared the topography of the SFSS weights with that of the FFSS and SFRS, the former of which being associated with assessing the similarity of faces other than one’s own, and the latter with recognizing one’s own face without explicitly assessing the degree of similarity. Both SFSS and FFSS responses significantly predicted similarity ratings obtained not only in their own condition, but also in the opposite condition, and their topography was visually very similar. PCA reconstruction of the commonalities between SFSS and FFSS showed that they are mainly related to the right and left regions of the EVC, which, given the above discussion, are heavily involved in faces similarity assessment. Regressing these commonalities out from the beta weights in the raw data yielded signatures that selectively predicted similarity scores in each condition but not in the other. These signatures showed no spatial overlap between both positive vs positive and negative vs negative weights. Most interestingly, overlaps in positive vs negative weights were found in the fusiform and orbital gyri (see Fig. 6 C). It can be hypothesized that, unlike EVC, which appear to be involved in assessing the similarity of faces regardless of the target to which similarity is being assessed, these regions are involved in target discrimination. In favor of this hypothesis is also the fact that these are the regions where the commonality between SFSS and SFRS is revealed (see Fig. 6 D). Thus, the right fusiform gyrus and the left OFC are the places where SFSS is similar with SFRS and is dissimilar with FFSS. From the data presented in Table 1 , it can be seen that both FFA and medial OFC (mOFC) are barely predictive of similarity scores. Perhaps this is not surprising in the case of mOFC, but it is puzzling that FFA, the core area of FSBA, is so weakly associated with subjective assessments of facial similarity. These results are consistent with the view that face similarity is first processed in EVC before being transferred to other areas of FSBA (Duchaine and Yovel 2016). As for FFA, it is involved in processing other aspects of the presented faces, such as whether to pay attention to them or ignore them (e.g., Gentile and Jansma 2010 ), or, in our case, whether it is my own or someone else’s face. As shown in Table 2 , compared to the other ROIs, the mOFC shows the least representative similarity to the behavioral data. Clearly, the mOFC is not a brain region specifically involved in face similarity assessment. Given the existing evidence of mOFC involvement in reward processing (Diekhof et al. 2012 ; Howard and Kahnt 2021 ) and frequently observed association of self-referential processing with activation of reward circuits (Chakraborty and Chakrabarti 2018 ; Devue et al. 2009 ; Ota and Nakano 2021 ; Platek et al. 2009 ; Zhan et al. 2016 ), one may speculate that self-recognition and self-face similarity assessment are more associated with reward (and, correspondingly, mOFC activation) than friend’s face similarity assessment. This partly is confirmed by the fact that mOFC ROI’s RDM was significantly associated with behavioral RDM in the ‘self’, but not in the ‘friend’ condition (see Table 2 ). Additionally, representation similarity of self-referential ROI was significantly higher in the ‘self’ than in the ‘friend’ condition, as shown by a paired samples permutation t-test across participants (t 52 = 2.05, p = 0.046). A possible limitation of this study is the relatively small sample size. However, a post hoc power calculation shows that with the effect sizes found in this study (Cohen’s d ranging from 2.58 to 13) a sample size of 54 participants achieves a power of 0.99. This is consistent with other MVPA studies generally showing high effect sizes for associations between brain signatures and mental states (e.g., Han et al. 2022 ; Kragel et al. 2018 ). 5. Conclusion To summarize, in this study, using MVPA approaches, we developed a brain signature of face similarity assessment and demonstrated its reliability in both intrapersonal and interpersonal domains in predicting subjective face similarity ratings. This signature response showed a perfect correlation with behavioral data and a large effect size in predicting the instantaneous (i.e., in each trial) similarity score for each participant. Moreover, it acted as a mediator between the degree of similarity objectively present in the image and the subjective assessment of similarity. Different analytical approaches including the traditional GLM, MVPA classification based on the whole brain, selected ROIs, and local neighborhood, and RSA converged in showing the primacy of early visual cortical areas (in particular, the right EVC) for face similarity assessment. Principal component analyses of three distinct signatures (SFSS, FFSS, and SFRS) allowed to separate brain regions associated with the assessment of face similarity independently of the target with which the similarity is judged and specifically associated with judging the similarity to one’s own face. It appears that the EVC is particularly involved in the first case whereas the fusiform gyrus and the OFC are the places where the distinction of targets is processed. The results of this analysis show the fruitfulness of considering brain signatures as representations of complex psychological processes, which can be analyzed by decomposing the signatures into their components and combining them in various ways to represent different combinations of relevant psychological states. This approach paves the way to what is considered the fundamental goal of cognitive neuroscience: establishing a correspondence between mind and brain (Kragel et al. 2018 ). Declarations Funding information This work was supported by budgetary funding for basic scientific research (theme No. 122042700001-9). CRediT authorship contribution statement G.G. Knyazev: Methodology, Conceptualization, Data analysis, Writing – original draft. A.N. Savostyanov: Data curation, Recruiting participants, Designing experimental setup, Writing – review and editing. A.V. Bocharov: Data curation, Writing – review and editing. A.E. Saprigyn: Data curation, Software, Designing experimental setup, Writing – review and editing. E.A. Levin: Data curation, Software, Writing – review and editing. D. A. Lebedkin: Data – collection and curation, Recruiting participants, Writing – review and editing. Declaration of interest None of the authors have a conflict of interest to disclose Data availability The data used to create the SFRS and the SFRS itself are available at https://doi.org/10.17605/OSF.IO/28HKV. Acknowledgements The study was supported by budgetary funding for basic scientific research (theme No. 122042700001-9). The authors thank Dmitrienko N.V. for help with data collection and Borisov S.V. for help with photography. MRI recordings were performed on the equipment of the International Tomography Center (ITC SB RAS). References Asakage S, Nakano T (2023) The salience network is activated during self‐recognition from both first‐person and third‐person perspectives. Hum Brain Mapp 44(2):559-570. https://doi.org/10.1002/hbm.26084 Åsberg Johnels J, Galazka MA, Sundqvist M, Hadjikhani N (2022) Left visual field bias during face perception aligns with individual differences in reading skills and is absent in dyslexia. Brit J Educ Psychol 00:1-10. https://doi.org/10.1111/bjep.12559 Baron RM, Kenny DA (1986) The Moderator-Mediator Variable Distinction in Social Psychological Research: Conceptual, Strategic, and Statistical Considerations. J Pers Soc Psychol 51(6):1173-1182. https://doi.org/10.1037/0022-3514.51.6.1173 Chakraborty A, Chakrabarti B (2018) Looking at my own face: Visual processing strategies in self–other face recognition. Front Psychol 9:121. https://doi.org/10.3389/fpsyg.2018.00121 Courville T, Thompson B (2001) Use of structure coefficients in published multiple regression articles: β is not enough. Educ Psychol Meas 61:229–248. http://epm.sagepub.com/cgi/content/abstract/61/2/229 Crosse MJ, Foxe JJ, Molholm S (2024) PERMUTOOLS: A MATLAB Package for Multivariate Permutation Testing. arXiv preprint arXiv:2401.09401. https://github.com/mickcrosse/PERMUTOOLS Devue C, Brédart S (2011) The neural correlates of visual self-recognition. Conscious Cogn 20(1):40-51. https://doi.org/10.1016/j.concog.2010.09.007 Devue C, Van der Stigchel S, Brédart S, Theeuwes J (2009) You do not find your own face faster; you just look at it longer. Cognition 111:114–122. https://doi.org/10.1016/j.cognition.2009.01.003 Diekhof EK, Kaps L, Falkai P, Gruber O (2012) The role of the human ventral striatum and the medial orbitofrontal cortex in the representation of reward magnitude–An activation likelihood estimation meta-analysis of neuroimaging studies of passive reward expectancy and outcome processing. Neuropsychologia 50(7):1252-1266. https://doi.org/10.1016/j.neuropsychologia.2012.02.007 Dimsdale-Zucker HR, Ranganath C (2018) Representational similarity analyses: a practical guide for functional MRI applications. In Handbook of behavioral neuroscience, Elsevier, Vol 28, pp 509-525. https://doi.org/10.1016/B978-0-12-812028-6.00027-6 Duchaine B, Yovel G (2015) A revised neural framework for face processing. Annu Rev Vis Sci 1(1):393-416. https://doi.org/10.1146/annurev-vision-082114-035518 Elfgren C, van Westen D, Passant U, Larsson EM, Mannfolk P, Fransson P (2006) fMRI activity in the medial temporal lobe during famous face processing. Neuroimage 30(2):609-616. https://doi.org/10.1016/j.neuroimage.2005.09.060 Gentile F, Jansma BM (2010) Neural competition through visual similarity in face selection. Brain Res 1351:172-184. https://doi.org/10.1016/j.brainres.2010.06.050 Gillihan SJ, Farah MJ (2005) Is self special? A critical review of evidence from experimental psychology and cognitive neuroscience. Psychol Bull 131(1):76–97. https://doi.org/10.1037/0033-2909.131.1.76 Han X, Ashar YK, Kragel P et al (2022) Effect sizes and test-retest reliability of the fMRI-based neurologic pain signature. Neuroimage 247:118844. https://doi.org/10.1016/j.neuroimage.2021.118844 Hansen F, DeBruine LM, Holzleitner IJ, Lee AJ, O'Shea KJ, Fasolt V (2020) Kin recognition and perceived facial similarity. J Vis 20(6):18-18. https://doi.org/10.1167/jov.20.6.18 Harrison MT, Strother L (2021) Does face-selective cortex show a left visual field bias for centrally-viewed faces?. Neuropsychologia 159:107956. https://doi.org/10.1016/j.neuropsychologia.2021.107956 Hassall C, Silver A, Turk DJ, Krigolson OE (2016) We are more selfish than we think: The endowment effect and reward processing within the human medial-frontal cortex. Q J Exp Psychol 69(6):1676-1686. https://doi.org/10.1080/17470218.2015.1091849 Haufe S, Meinecke F, Görgen K, Dähne S, Haynes JD, Blankertz B, Bießmann F (2014) On the interpretation of weight vectors of linear models in multivariate neuroimaging. NeuroImage 87:96-110. https://doi.org/10.1016/j.neuroimage.2013.10.067 Haxby JV, Hoffman EA, Gobbini MI (2000) The distributed human neural system for face perception. Trends Cogn Sci 4(6):223-233. https://doi.org/10.1016/S1364-6613(00)01482-0 Haynes JD, Rees G (2006) Decoding mental states from brain activity in humans. Nat Rev Neurosci 7:523-534. https://doi.org/10.1038/nrn1931 Hougaard A, Jensen BH, Amin FM, Rostrup E, Hoffmann MB, Ashina M (2015) Cerebral Asymmetry of fMRI-BOLD Responses to Visual Stimulation. PLoS ONE 10(5):e0126477. https://doi.org/10.1371/journal.pone.0126477 Howard JD, Kahnt T (2021) To be specific: The role of orbitofrontal cortex in signaling reward identity. Behav Neurosci 135(2):210-217. https://doi.org/10.1037/bne0000455 Hu C, Di X, Eickhoff SB, Zhang M, Peng K, Guo H, Sui J. (2016) Distinct and common aspects of physical and psychological self-representation in the brain: A meta-analysis of self-bias in facial and self-referential judgements. Neurosci Biobehav Rev 61:197-207. https://doi.org/10.1016/j.neubiorev.2015.12.003 Jabakhanji R, Vigotsky AD, Bielefeld J, Huang L, Baliki MN, Iannetti G, Apkarian AV (2022) Limits of decoding mental states with fMRI. Cortex 149:101-122. https://doi.org/10.1016/j.cortex.2021.12.015 Janowska A, Balugas B, Pardillo M et al (2021) The Neurological Asymmetry of Self-Face Recognition. Symmetry 13(7):1135. https://doi.org/10.3390/sym13071135 Kanwisher N, McDermott J, Chun MM (1997) The fusiform face area: a module in human extrastriate cortex specialized for face perception. J Neurophysiol 17:4302-4311. https://doi.org/10.1523/JNEUROSCI.17-11-04302.1997 Knyazev GG, Savostyanov AN, Bocharov AV, Saprigyn AE, Levin EA (2024) Investigating the properties of fMRI-based signature of recognizing one’s own face. Biol Psychol 108960. https://doi.org/10.1016/j.biopsycho.2024.108960 Koo TK, Li MY (2016) A guideline of selecting and reporting intraclass correlation coefficients for reliability research. J Chiropr Med 15:155–163. https://doi.org/10.1016/j.jcm.2016.02.012 Kragel PA, Koban L, Barrett LF, Wager TD (2018) Representation, pattern information, and brain signatures: from neurons to neuroimaging. Neuron 99:257–273. https://doi.org/10.1016/j.neuron.2018.06.009 Kriegeskorte N, Mur M, Bandettini PA (2008) Representational similarity analysis-connecting the branches of systems neuroscience. Front Syst Neurosci 2:249. https://doi.org/10.3389/neuro.06.004.2008 Levine SC, Banich MT, Koch-Weser MP (1988) Face recognition: a general or specific right hemisphere capacity?. Brain Cogn 8(3):303-325. https://doi.org/10.1016/0278-2626(88)90057-7 Lyons M, Morikawa K (2020) A Linked Aggregate Code for Processing Faces (Revised Version). arXiv preprint arXiv:2009.08281. https://doi.org/10.48550/arXiv.2009.08281 Molnar-Szakacs I, Uddin LQ (2023) Laterality and hemispheric specialization of self-face recognition. Neuropsychologia 108586. https://doi.org/10.1016/j.neuropsychologia.2023.108586 Morita T, Tanabe HC, Sasaki AT, Shimada K, Kakigi R, Sadato N (2014) The anterior insular and anterior cingulate cortices in emotional processing for self-face recognition. Soc Cogn Affect Neurosci 9(5):570-579. https://doi.org/10.1093/scan/nst011 Nili H, Wingfield C, Walther A, Su L, Marslen-Wilson W, Kriegeskorte N (2014) A toolbox for representational similarity analysis. PLoS Comput Biol 10(4):e1003553. https://doi.org/10.1371/journal.pcbi.1003553 Norman KA, Polyn SM, Detre GJ, Haxby JV (2006) Beyond mind-reading: multi-voxel pattern analysis of fMRT data. Trends Cogn Sci 10, 424-430. https://doi.org/10.1016/j.tics.2006.07.005 Northoff G, Heinzel A, de Greck M, Bermpohl F, Dobrowolny H, Panksepp J (2006) Self-referential processing in our brain - a meta-analysis of imaging studies on the self. NeuroImage 31(1):440–457. https://doi.org/10.1016/j.neuroimage.2005.12.002 Northoff G, Qin P, Feinberg TE (2011) Brain imaging of the self – conceptual, anatomical and methodological issues. Conscious Cogn 20(1):52–63. https://doi.org/10.1016/j.concog.2010.09.011 Oosterhof NN, Connolly AC, Haxby JV (2016) CoSMoMVPA: multi-modal multivariate pattern analysis of neuroimaging data in Matlab/GNU Octave. Front Neuroinform 10:27. https://doi.org/10.3389/fninf.2016.00027 Ota C, Nakano T (2021) Self-face activates the dopamine reward pathway without awareness. Cereb Cortex 31(10):4420-4426. https://doi.org/10.1093/cercor/bhab096 Pereira F, Mitchell T, Botvinick M (2009) Machine learning classifiers and fMRT: a tutorial overview. NeuroImage 45:199-209. https://doi.org/10.1016/j.neuroimage.2008.11.007 Platek SM, Krill AL, Wilson B (2009) Implicit trustworthiness ratings of self-resembling faces activate brain centers involved in reward. Neuropsychologia 47(1):289-293. https://doi.org/10.1016/j.neuropsychologia.2008.07.018 Popal H, Wang Y, Olson IR (2019) A guide to representational similarity analysis for social neuroscience. Soc Cogn Affect Neurosci 14(11):1243-1253. https://doi.org/10.1093/scan/nsz099 Pourtois G, Schwartz S, Seghier ML, Lazeyras, Vuilleumier P (2005) View-independent coding of face identity in frontal and temporal cortices is modulated by familiarity: an event-related fMRI study. Neuroimage 24(4):1214-1224. https://doi.org/10.1016/j.neuroimage.2004.10.038 Prince JS, Charest I, Kurzawski JW, Pyles JA, Tarr MJ, Kay KN (2022) Improving the accuracy of single-trial fMRI response estimates using GLMsingle. Elife 11:e77599. https://doi.org/10.7554/eLife.77599 Ramon M, Vizioli L, Liu-Shuang J, Rossion B (2015) Neural microgenesis of personally familiar face recognition. PNAS 112(35):E4835-E4844. https://doi.org/10.1073/pnas.1414929112 Ritchie JB, Kaplan DM, Klein C (2019) Decoding the brain: Neural representation and the limits of multivariate pattern analysis in cognitive neuroscience. Brit J Philos Sci 70:581–607. https://doi.org/10.1093/bjps/axx023 Rossion B (2014) Understanding face perception by means of prosopagnosia and neuroimaging. Front Biosci 6:258–307. https://doi.org/10.2741/e706 Shrout PE, Bolger N (2002) Mediation in experimental and nonexperimental studies: New procedures and recommendations. Psychol Methods 7:422-445. https://doi.org/10.1037/1082-989X.7.4.422 Sugiura M, Miyauchi CM, Kotozaki Y et al (2015) Neural mechanism for mirrored self-face recognition. Cereb Cortex 25(9):2806-2814. https://doi.org/10.1093/cercor/bhu077 Sui J, Humphreys GW (2015) The interaction between self-bias and reward: Evidence for common and distinct processes. Q J Exp Psychol 68(10):1952-1964. https://doi.org/10.1080/17470218.2015.1023207 Tramacere A (2022) Face yourself: The social neuroscience of mirror gazing. Front Psychol 13:949211. https://doi.org/10.3389/fpsyg.2022.949211 Uddin LQ, Kaplan JT, Molnar-Szakacs I, Zaidel E, Iacoboni M (2005) Self-face recognition activates a frontoparietal “mirror” network in the right hemisphere: an event-related fMRI study. NeuroImage 25(3):926-935. https://doi.org/10.1016/j.neuroimage.2004.12.018 Vigotsky AD, Iannetti GD, Apkarian AV (2024) Mental state decoders: game-changers or wishful thinking?. Trends Cogn Sci 28(10):884-895. https://doi.org/10.1016/j.tics.2024.06.004 Wager TD, Atlas LY, Leotti LA, Rilling JK (2011) Predicting individual differences in placebo analgesia: contributions of brain activity during anticipation and pain experience. J Neurosci 31:439-452. https://doi.org/10.1523/JNEUROSCI.3420-10.2011 Wager TD, Atlas LY, Lindquist MA, Roy M, Woo CW, Kross E (2013) An fMRI-based neurologic signature of physical pain. N Engl J Med 368(15):1388-1397. https://doi.org/10.1056/NEJMoa1204471 Wager TD, Waugh CE, Lindquist M, Noll DC, Fredrickson BL, Taylor SF (2009) Brain mediators of cardiovascular responses to social threat, Part I: Reciprocal dorsal and ventral sub-regions of the medial prefrontal cortex and heart-rate reactivity. NeuroImage 47:821-835. https://doi.org/10.1016/j.neuroimage.2009.05.043 Weiskopf DA (2021) Data Mining the Brain to Decode the Mind. In: Calzavarini F, Viola M (eds) Neural Mechanisms. Studies in Brain and Mind, vol 17. Springer, Cham, pp 85-110. https://doi.org/10.1007/978-3-030-54092-0_5 Worsley KJ, Marrett S, Neelin P, Vanda AC, Friston KJ, Evans AC (1996) A unified statistical approach or determining significant signals in images of cerebral activation. Hum Brain Mapp 4:58-73. https://doi.org/10.1002/(SICI)1097-0193(1996)4:13.0.CO;2-O Yarkoni T, Poldrack RA, Nichols TE, Van Essen DC, Wager TD (2011) Large-scale automated synthesis of human functional neuroimaging data. Nat methods 8(8):665-670. https://doi.org/10.1038/nmeth.1635 Yue X, Biederman I, Mangini MC, von der Malsburg C, Amir O (2012) Predicting the psychophysical similarity of faces and non-face complex shapes by image-based measures. Vis Res 55:41-46. https://doi.org/10.1016/j.visres.2011.12.012 Zhan Y, Chen J, Xiao X, Li J, Yang Z, Fan W, Zhong Y (2016) Reward promotes self-face processing: An event-related potential study. Front Psychol 7:735. https://doi.org/10.3389/fpsyg.2016.00735 Zhou F, Zhao W, Qi Z et al (2021) A distributed fMRI-based signature for the subjective experience of fear. Nat Commun 12(1):6643. https://doi.org/10.1038/s41467-021-26977-3 Żochowska A, Nowicka MM, Wójcik MJ, Nowicka A (2021) Self-face and emotional faces—are they alike?. Soc Cogn Affect Neurosci 16(6):593-607. https://doi.org/10.1093/scan/nsab020 Tables Table 1. Prediction-outcome correlations for ROI-based signatures in the ‘self’ and ‘friend’ conditions. ROI ‘self’ ‘friend’ Early visual 0.69 0.66 Inferior occipital 0.34 0.32 Fusiform face area 0.04 0.05 Face recognition 0.18 0.16 Self-referential 0.22 0.20 Medial orbitofrontal 0.06 0.09 All correlations are significant at p < 0.05, FDR-corrected. Table 2. RSA results. Spearman correlations between ROI RDMs and behavioral RDMs averaged across participants. Significance of correlations was tested using a one-sample permutation t-test across participants. ROI ‘self’ ‘friend’ Early visual 0.64* 0.62* Inferior occipital 0.22* 0.23* Fusiform face area 0.21* 0.20* Face recognition 0.14* 0.12* Self-referential 0.25* 0.13* Medial orbitofrontal 0.08* 0.001 * correlation is significant at p < 0.05, FDR-corrected. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7570992","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":531835637,"identity":"a50af478-6a02-42dd-affb-87a1597f1612","order_by":0,"name":"Gennady Knyazev","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA30lEQVRIiWNgGAWjYBACCTiLvYFkLTwHSNYikUCkFsnZPYYff1Qctpsf+fbgY949Nnm6DbzPJPBpkZY5YywhceZw8sbbecnGPM/Sis0OsJvh1SInkWMgYdiWlmw4O8dMmufA4cRtB9jYCGkx/pH4D6hl5hnz30RpkZbIMZM42GBjJy/BY8ZMlBbJGWlllg3HbBIMeHKMJeccSEvcdpiN2QKfFokbyZtv/qiRsJdvP2P44c0Bm8Rtx9sYb+DTwsDAYQAiEzccgAkw41cPBOwPQKS9fANBlaNgFIyCUTBSAQD7Y0e3jIIOnwAAAABJRU5ErkJggg==","orcid":"","institution":"Scientific Research Institute of Neurosciences and Medicine","correspondingAuthor":true,"prefix":"","firstName":"Gennady","middleName":"","lastName":"Knyazev","suffix":""},{"id":531835638,"identity":"cb0a38b1-ab55-41ce-8ab0-246b93f8ee13","order_by":1,"name":"Alexander Savostyanov","email":"","orcid":"","institution":"Scientific Research Institute of Neurosciences and Medicine","correspondingAuthor":false,"prefix":"","firstName":"Alexander","middleName":"","lastName":"Savostyanov","suffix":""},{"id":531835639,"identity":"9e8cfe8d-9919-49a8-8b89-f444846cfccf","order_by":2,"name":"Andrey Bocharov","email":"","orcid":"","institution":"Scientific Research Institute of Neurosciences and Medicine","correspondingAuthor":false,"prefix":"","firstName":"Andrey","middleName":"","lastName":"Bocharov","suffix":""},{"id":531835640,"identity":"4c320884-c850-4671-8acd-7ec14bbd07fc","order_by":3,"name":"Alexander Saprigyn","email":"","orcid":"","institution":"Scientific Research Institute of Neurosciences and Medicine","correspondingAuthor":false,"prefix":"","firstName":"Alexander","middleName":"","lastName":"Saprigyn","suffix":""},{"id":531835641,"identity":"98dbb437-e5c3-480e-9143-2be615f0e71b","order_by":4,"name":"Evgeny Levin","email":"","orcid":"","institution":"Scientific Research Institute of Neurosciences and Medicine","correspondingAuthor":false,"prefix":"","firstName":"Evgeny","middleName":"","lastName":"Levin","suffix":""},{"id":531835642,"identity":"5b17b213-556c-4f33-a050-f51fb21455ff","order_by":5,"name":"Dmitri Lebedkin","email":"","orcid":"","institution":"Scientific Research Institute of Neurosciences and Medicine","correspondingAuthor":false,"prefix":"","firstName":"Dmitri","middleName":"","lastName":"Lebedkin","suffix":""}],"badges":[],"createdAt":"2025-09-09 07:53:47","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7570992/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7570992/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":94047620,"identity":"7421e887-1b58-4bf2-a086-73728d613f8c","added_by":"auto","created_at":"2025-10-21 23:15:54","extension":"tif","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":85740,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.1.tif","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/784dfd7a454aaff24402d18c.tif"},{"id":94046665,"identity":"21451d9c-0aef-418b-a2fd-ca71dfd66e51","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"tif","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2367446,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.2.tif","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/5927cc26fc71299a47266cac.tif"},{"id":94046659,"identity":"59a469be-e807-4fa8-8ad2-1e932eb7662a","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":70039,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/516f7d17e8a83556f47a7731.docx"},{"id":94046663,"identity":"d8ef98e8-b978-4c3e-a0d1-688623a77fbd","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"tif","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":8410604,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.3.tif","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/5488d326fe19eb64bfcfc370.tif"},{"id":94046675,"identity":"8ada4bf8-8abc-4056-93b7-871a08421db8","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"tif","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":612528,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.4.tif","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/f92a780fc5256ac6c5a9dfdb.tif"},{"id":94046668,"identity":"b772267a-16b3-486f-9942-b492bbbda181","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"tif","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1461510,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.5.tif","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/bf51228abec8a2c91a20d393.tif"},{"id":94046681,"identity":"e68e7646-ccb5-424b-bd3e-4590947dc61d","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"tif","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1463476,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.6.tif","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/aece9ab5fdd69135d51257ed.tif"},{"id":94047628,"identity":"005631e0-425b-48dc-982c-18aee45f00d7","added_by":"auto","created_at":"2025-10-21 23:15:54","extension":"tif","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5476398,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.7.tif","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/86b0c83572c4e66f4921236a.tif"},{"id":94047623,"identity":"6cf483f8-ca6e-42bd-8c58-d022d81b4b29","added_by":"auto","created_at":"2025-10-21 23:15:54","extension":"json","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":8122,"visible":true,"origin":"","legend":"","description":"","filename":"6b3d100218ca4a1bbecc3756bcfaee20.json","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/30a3fbf0587fadfa6f584432.json"},{"id":94046677,"identity":"f1a35812-5592-473a-a19e-7194170ed3b8","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"xml","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":174253,"visible":true,"origin":"","legend":"","description":"","filename":"6b3d100218ca4a1bbecc3756bcfaee201enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/973b8c6712f421f57bcbcf3f.xml"},{"id":94046662,"identity":"09ef4ecb-ca65-4f73-8dbc-37930b7ab4b3","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"tif","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":85740,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.1.tif","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/82040ddc0c1afeaefdcc3e27.tif"},{"id":94047627,"identity":"7a6e7958-c1ad-4a8c-a57f-fc3fd4e143e9","added_by":"auto","created_at":"2025-10-21 23:15:54","extension":"tif","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2367446,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.2.tif","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/b4f9e7c9efd05a5a528628c6.tif"},{"id":94046671,"identity":"82598583-1942-452a-b7fe-ff17a7a17ea9","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"tif","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":8410604,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.3.tif","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/8e389823afe216230388df8f.tif"},{"id":94046682,"identity":"f7324fbd-3540-45f5-93d4-5ca8875b4703","added_by":"auto","created_at":"2025-10-21 23:07:55","extension":"tif","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":612528,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.4.tif","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/e4b411d42ab665465d633edd.tif"},{"id":94047629,"identity":"c98ff07f-1be1-43bc-a55c-a14a157d4fd1","added_by":"auto","created_at":"2025-10-21 23:15:55","extension":"tif","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1461510,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.5.tif","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/c6c0a351a74f4b7fb02bc4fb.tif"},{"id":94047624,"identity":"9c5badb1-0146-4bc2-a20d-c94381d88d2d","added_by":"auto","created_at":"2025-10-21 23:15:54","extension":"tif","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1463476,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.6.tif","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/0050f51a3e77f4946fc2a5a2.tif"},{"id":94046687,"identity":"6c832fa6-8a70-4850-83bb-5599b9865ed8","added_by":"auto","created_at":"2025-10-21 23:07:55","extension":"tif","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5476398,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.7.tif","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/ad4022ce87921fd7e1d39038.tif"},{"id":94046666,"identity":"5cff740b-4b6b-4c00-ad32-ae66bdb7a773","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":14077,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig.1.png","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/81e99e0ad0811aa3c95d4615.png"},{"id":94046672,"identity":"34f93cee-f12d-449b-842a-b84ef72a3552","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":176141,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig.2.png","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/e5f1f6b12be83a6eb484c193.png"},{"id":94046686,"identity":"de35bb0f-7b50-4fbd-b071-8939804e9fb2","added_by":"auto","created_at":"2025-10-21 23:07:55","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":623858,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig.3.png","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/d51858b02b7663495c5db2ca.png"},{"id":94046678,"identity":"9b2b8725-6b16-4d78-a67f-75e2210f601c","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":79488,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig.4.png","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/e782844b388be280b54c8aac.png"},{"id":94046674,"identity":"0c994b7e-dcbb-4d3f-a03d-52db906b09cf","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":246112,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig.5.png","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/155c43cd2be8389c306b5e07.png"},{"id":94047626,"identity":"4ef2c8ea-2c91-4f28-b95d-13f41ec7afa6","added_by":"auto","created_at":"2025-10-21 23:15:54","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":97362,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig.6.png","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/05dd85d6d87a6b77ad07f2a8.png"},{"id":94047625,"identity":"f1fa014c-6262-4420-ac9f-5fb6e24554eb","added_by":"auto","created_at":"2025-10-21 23:15:54","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":438513,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig.7.png","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/b3e46a91960fe9c750c1424f.png"},{"id":94046685,"identity":"8caa8895-d650-46bf-98f0-5333628498bc","added_by":"auto","created_at":"2025-10-21 23:07:55","extension":"xml","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":173443,"visible":true,"origin":"","legend":"","description":"","filename":"6b3d100218ca4a1bbecc3756bcfaee201structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/7ee3b93fb273c9fa09a75ae8.xml"},{"id":94046684,"identity":"c16a5f67-2874-4bda-8e2b-ff2fb9e0f3c0","added_by":"auto","created_at":"2025-10-21 23:07:55","extension":"html","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":184159,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/42b7d659250f3377c3cb326a.html"},{"id":94046655,"identity":"0c755c9f-f218-42c9-bb6c-ad69ab7b44b3","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":35479,"visible":true,"origin":"","legend":"\u003cp\u003eBehavioral data. An interaction between similarity and familiarity. X axis – morphing stages. Y axis – estimated marginal means of similarity ratings. Blue line – a familiar opposite face; green line – an unfamiliar opposite face\u003c/p\u003e","description":"","filename":"Fig.1.png","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/d685bea3492909107326eaf1.png"},{"id":94048521,"identity":"478f04b6-55f2-47ae-a980-10e0b4788715","added_by":"auto","created_at":"2025-10-21 23:23:54","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":593934,"visible":true,"origin":"","legend":"\u003cp\u003eUnivariate GLM analysis. Main effect of similarity. A – the most significant effects are observed in the right (RLG) and left (LLG) lingual gyrus. The color scale shows F values. B - contrast estimates and 90% confidence intervals of the LLG and RLG response at each similarity rating point. The RLG shows a gradual increase in activation with increasing similarity ratings, while the LLG shows the opposite dynamics\u003c/p\u003e","description":"","filename":"Fig.2.png","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/b40dea7fd85868042ad40a9c.png"},{"id":94046660,"identity":"126bdd32-29c5-44f6-b9d8-c16599c7e90f","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":2880767,"visible":true,"origin":"","legend":"\u003cp\u003eSelf- and friend-face similarity signature maps and activation patterns converted into z-scores and FDR-thresholded at q \u0026lt; 0.05. A – SFSS; B – FFSS; C – self-face similarity activation pattern; D – friend-face similarity activation pattern. The warm colors indicate positive weights and cool colors indicate negative weights. Only clusters \u0026gt; 100 voxels are shown. E – the overlap between the SFSS predictive weights map and the activation pattern map; F – the overlap between the FFSS predictive weights map and the activation pattern map. Only clusters \u0026gt; 10 voxels are shown\u003c/p\u003e","description":"","filename":"Fig.3.png","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/d41c8042cc2d88cd994c5c8b.png"},{"id":94046656,"identity":"6ee81437-4c4b-4140-8b20-511bee2840b5","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":132820,"visible":true,"origin":"","legend":"\u003cp\u003eModel performance evaluated as increasing numbers of voxels (x axis) were used to predict the similarity ratings. The y axis denotes the cross-validated prediction-outcome correlation. The dots indicate the averaged over 100 iterations correlation coefficients, solid line indicates the mean parametric fit\u003c/p\u003e","description":"","filename":"Fig.4.png","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/d84c6c3a6847fe45aa60e1e8.png"},{"id":94046661,"identity":"5259cef8-4569-4e33-8c90-d3af100ce773","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":709868,"visible":true,"origin":"","legend":"\u003cp\u003eResults of bootstrap tests of the mediation model slopes: \u003cstrong\u003ea\u003c/strong\u003e - X→M (degree of morphing→SFSS response); \u003cstrong\u003eb\u003c/strong\u003e - M→Y (SFSS response→similarity scores); \u003cstrong\u003ec’\u003c/strong\u003e - X→Y controlling M (degree of morphing→similarity scores controlling for signature response); \u003cstrong\u003ec\u003c/strong\u003e - X→Y without M controlling (degree of morphing→similarity scores without controlling for signature response)\u003c/p\u003e","description":"","filename":"Fig.5.png","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/3cdb8965fcde7842d978de98.png"},{"id":94046676,"identity":"5669a3ce-f02c-46cb-9714-7fe976821b7f","added_by":"auto","created_at":"2025-10-21 23:07:54","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":392703,"visible":true,"origin":"","legend":"\u003cp\u003eSearchlight regression analysis. For each searchlight volume, prediction accuracy for each participant was calculated as the correlation between predicted and actual similarity scores. Correlations that were significant across participants at p \u0026lt; 0.001 were assigned to the central voxel of the searchlight volume. The warm colors indicate positive correlation coefficients. Only clusters \u0026gt; 100 voxels are shown.\u003c/p\u003e","description":"","filename":"Fig.6.png","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/8d64d04f584812e1d67d7464.png"},{"id":94047622,"identity":"4e84cbf4-c848-45f5-9a23-b455b861be47","added_by":"auto","created_at":"2025-10-21 23:15:54","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":1731851,"visible":true,"origin":"","legend":"\u003cp\u003eDecomposing SFSS and FFSS into common and distinct components. 6A and 6B – results of PCA of SFSS and FFSS weights. A - FDR-thresholded map of reconstructed pattern (REC) after extracting the first component; B - FDR-thresholded map of residual pattern after extracting the first component. 6C - the overlap between positive vs negative weights of PSFSS and PFFSS. 6D – FDR-thresholded map of reconstructed pattern of commonalities between SFSS and SFRS. The warm colors indicate positive weights and cool colors indicate negative weights\u003c/p\u003e","description":"","filename":"Fig.7.png","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/d82bb3d5eedfbb3686b3f6bf.png"},{"id":102746084,"identity":"bf47ab95-1fb2-410d-a863-c43477a8bbfc","added_by":"auto","created_at":"2026-02-16 08:55:37","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":10451594,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7570992/v1/7dcbbb6f-c23c-4029-bef9-e4cd6eba8bbb.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Neural Signature of Similarity Assessment between a Face and One’s Own Face","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eFace similarity assessment is one of the basic social skills that greatly influences the social behavior of an individual. In particular, facial similarity to one\u0026rsquo;s own face is a factor that may indirectly influence the judgment of a person\u0026rsquo;s trustworthiness (Hansen et al \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Assessing facial similarity to one\u0026rsquo;s own face is akin to self-recognition, the brain underpinning of which have been the subject of many recent studies (Asakage and Nakano \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Devue and Br\u0026eacute;dart \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Hu et al. \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Janowska et al. \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Knyazev et al. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Molnar-Szakacs and Uddin \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Morita et al. \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Ramon et al. \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Sugiura et al. \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Tramacere \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Żochowska et al. \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Nevertheless, assessing facial similarity to one\u0026rsquo;s own face and recognizing one\u0026rsquo;s own face are conceptually different tasks. Therefore, it can be expected that the neural signatures of these tasks may partially overlap and partially diverge. The present study aimed to explore using functional magnetic resonance imaging (fMRI) the brain signature of the assessment of facial similarity to one\u0026rsquo;s own face.\u003c/p\u003e\u003cp\u003eThe process of assessing the similarity of a face to one\u0026rsquo;s own face has two aspects. The first is the assessment of the similarity of faces regardless of the object with which the similarity is assessed. The second aspect is associated specifically with the assessment of the similarity of a face to one\u0026rsquo;s own face. To distinguish between these two aspects, we included two tasks in the experimental design. These tasks used the same set of morphed images, but in one case the participant\u0026rsquo;s task was to evaluate the similarity to his or her own face and in the other case to the face of another person.\u003c/p\u003e\u003cp\u003eSince the pioneering work of Kanwisher et al. (\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e1997\u003c/span\u003e), it is generally accepted that face processing in the human brain is mainly associated with face-selective brain areas (FSBA), such as the fusiform face area (FFA), the occipital face area (OFA), and the posterior part of the superior temporal sulcus (pSTS-FA). Haxby et al. (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2000\u003c/span\u003e) proposed a highly influential neurocognitive model of face processing, according to which OFA implements early stages of face processing and directs its output to FFA, which represents unchanging aspects of the face, and to pSTS-FA, which represents changing aspects of the face. The information from this network is further processed in the extended face network. Later findings (reviewed in Duchaine and Yovel 2016) prompted a revision of this model. In particular, these findings indicate that multiple pathways directly connect early visual cortex (EVC) with multiple face-selective regions.\u003c/p\u003e\u003cp\u003eAn important feature of face processing in the brain is the dominance of the right hemisphere in face perception. Superior behavioral performance in the left visual hemifield (the so-called left visual field bias) and a preferential activation of right FSBA during face processing have been reported repeatedly (\u0026Aring;sberg Johnels et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Harrison and Strother \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Hougaard et al. \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Levine et al. \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e1988\u003c/span\u003e). Given this information, one would expect that the assessment of facial similarity in general should be reflected primarily in right-sided changes in EVC and FSBA.\u003c/p\u003e\u003cp\u003eWhen it comes to the second aspect, that is, the assessment of the similarity of a face specifically to one\u0026rsquo;s own face, it can be assumed to be akin to various types of self-referential processing, the neural basis of which has been the subject of numerous neuroimaging studies (e.g., Northoff et al. \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2006\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). It is often emphasized that there are at least two aspects of the self, the physical self and the psychological self (Gillihan and Farah \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2005\u003c/span\u003e). Recognizing one\u0026rsquo;s own face is a paradigmatic example of processing the physical self. Meta-analyses of neuroimaging studies have shown that these two aspects of self-processing engage both distinct and shared brain regions, the latter including the dorsal anterior cingulate cortex (ACC) and the left inferior frontal gyrus (IFG), extending to the insula (Hu et al. \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). An important feature of self-processing is that it is frequently associated with the activation of reward circuits (e.g., Hassall et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Sui and Humphreys \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). This is particularly true for self-face processing (Chakraborty and Chakrabarti \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Devue et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Ota and Nakano \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Platek et al. \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Zhan et al. \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Thus, one might expect that assessing similarity specifically to one\u0026rsquo;s own face should engage brain regions associated with self-referential processing, such as the ACC, IFG, and reward circuits, such as the orbitofrontal cortex (OFC, see e.g., Diekhof et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Howard and Kahnt \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eOne of the most important questions of such kind of research is the approach to data analysis. The mass univariate approach (Worsley et al. \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e1996\u003c/span\u003e), which has dominated the field in recent decades and has been used in the vast majority of published studies, has been criticized (e.g., Zhou et al. \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) and has recently tended to be supplanted by various variants of multivariate pattern analysis (MVPA) (Haynes and Rees, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Norman et al. \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Pereira et al. \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). In the study of brain-mental state associations, MVPA methods yield much greater effect sizes than the traditional mass univariate methods and good test-retest reliability (Kragel et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). However, these methods also have their limitations and should be used with caution (Haufe et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Jabakhanji et al. \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Ritchie et al. \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Weiskopf \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Vigotsky et al. \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). In this study, we used both the traditional and MVPA methods to analyze fMRI data recorded during presentation of self-to-other morphed face images (e.g., Uddin et al. \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2005\u003c/span\u003e), where the other was a familiar or unfamiliar person. To identify spatial patterns of brain activity associated with face similarity assessment, we used machine learning classification based on whole brain volume, selected regions of interest (ROIs), or local neighborhood as implemented in the searchlight approach. We also used the representational similarity analysis (RSA) (Kriegeskorte et al. \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2008\u003c/span\u003e) to reveal where in the brain the similarity between spatial patterns of activation matched the similarity between individual ratings of facial similarity.\u003c/p\u003e\u003cp\u003eThe brain\u0026rsquo;s response to viewed images can be seen as a mediator between the proportion of the self that is actually present in the image and the assessment of its similarity to one\u0026rsquo;s own face. Indeed, the binary decision (self or not-self) has been shown to be strongly correlated with the actual similarity of the image to one\u0026rsquo;s own face and partially mediated by the brain\u0026rsquo;s graded response to the proportions of self that are present in the images (Knyazev et al. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). In this study, we aimed to test whether the signature response acts as a mediator between the degree of image deviation from one\u0026rsquo;s own face and the rating of its similarity to one\u0026rsquo;s own face.\u003c/p\u003e\u003cp\u003eThe most relevant issues in the study of brain signatures (or \u0026lsquo;decoders\u0026rsquo;) of specific mental states or psychological processes are the consistency and reliability of the identified spatial patterns and their relationship to the brain activity associated with these states. To answer the first question, we compared the topography of brain maps created using traditional mass univariate and different MVPA methods, and tested the reliability of signature expression across participants and on retesting. The traditional and MVPA methods are based on different ideologies. The former implicitly assume a functional organization of the brain, implying that compactly localized activations in specific \u0026lsquo;functional\u0026rsquo; centers are sufficient to explain the brain\u0026rsquo;s involvement in stimulus processing (Worsley et al. \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e1996\u003c/span\u003e). The latter suggest that the content of a mental state is \u0026lsquo;encoded\u0026rsquo; in a broadly spatially distributed pattern of neural activity, and the most accurate prediction of relevant states can only be achieved by accounting for this broad distribution (Zhou et al. \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). To address this issue, we compared the predictive performance of decoders based on the whole brain volume with that of decoders based on specific functionally relevant ROIs.\u003c/p\u003e\u003cp\u003eThe second question is more difficult to answer. Some authors argue that although decoders based on machine learning can reliably predict the mental states on which they have been trained, their relevance to brain activity remains questionable (Haufe et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Jabakhanji et al. \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Ritchie et al. \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Weiskopf \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Vigotsky et al. \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). It is argued that machine learning classification methods can pick up information that is useful for improving prediction accuracy but is irrelevant to the task at hand (Haufe et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Others argue that the machine learning used to create decoders is not needed at all, since the predictive performance of these decoders is no better than that of the mass-univariate encoders used to train them (Jabakhanji et al. \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). To address these issues, we compared the predictive performance of obtained decoders with that of mass-univariate encoders, as proposed by Jabakhanji et al. (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), and by comparing its weights with \u0026lsquo;activation patterns\u0026rsquo;, as proposed by Haufe et al. (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). It is also argued that the information extracted by machine learning algorithms is highly redundant and more or less evenly distributed throughout the brain (Jabakhanji et al. \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). To test this, we examined the effect of spatial smoothing on prediction performance and tested how the number of voxels affects the effect size of prediction-outcome association, and whether randomly selected voxels can demonstrate prediction performance comparable to whole-brain prediction.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1. Participants\u003c/h2\u003e\u003cp\u003eThe sample used to develop the self-face similarity signature (SFSS) consisted of 54 right-handed participants (28 women, mean age 22 years, SD\u0026thinsp;=\u0026thinsp;2.4). To compare the SFSS with the self-face recognition signature (SFRS) we used the data described in (Knyazev et al. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). This sample consisted of 44 right-handed participants (26 women, mean age 26.5 years, SD\u0026thinsp;=\u0026thinsp;9). Exclusion criteria in both cases were serious medical conditions, history of epilepsy, substance abuse or dependence, and all contraindications to fMRI. The study complied with the Declaration of Helsinki of the World Medical Association and was approved by the Ethics Committee of the Institute of Neurosciences and Medicine. All participants gave written informed consent to participate in the experiment.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2. Face similarity judgement experimental procedure\u003c/h2\u003e\u003cp\u003eEach participant was invited into the laboratory with a good acquaintance or friend of the same gender. In addition, another person of the same age and gender, unfamiliar to the participant, was invited. All participants were photographed. The photographs were taken in frontal projection and then a mask was applied to crop out all hair details. Before morphing, point distinctive features such as birthmarks or blemishes were removed from photographs. Morphing software (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://alyssaq.github.io/face_morpher/\u003c/span\u003e\u003cspan address=\"http://alyssaq.github.io/face_morpher/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) was used to create intermediate images for all pairwise combinations (i.e., me vs friend, me vs stranger, and friend vs stranger). Images were ranked on a scale from 0 to 5, where 5 is the target face (i.e. the face to which similarity is being assessed) and 0 is the face of another person without morphing. The intermediate gradations are the different stages of morphing. Participants were given the following instructions: \u0026ldquo;In this experiment, you will be shown pictures of faces and asked to rate the degree of similarity of each face to the target face.\u0026rdquo; The experiment consisted of two sessions. In the first session, the target was the participant\u0026rsquo;s own face, and in the second session - the face of his\\her friend. The order of sessions was chosen at random for each participant. Each session consisted of three runs separated by two-minute rest intervals. Each run involved the presentation of two types of morphed images, i.e., morphed toward familiar (i.e., participant\u0026rsquo;s own or his\\her friend\u0026rsquo;s face) or unfamiliar (taken from the database) face. The images were presented in random order. First, a cross appeared in the center of the screen, then after 1\u0026thinsp;\u0026plusmn;\u0026thinsp;0.1 s - an image of a face, which remained on the screen for 2.5 s, after which the scale appeared, which allowed to select the degree of similarity. The scale ranged from 0 (not the target) to 10 (exactly the target). In data analysis, participants\u0026rsquo; ratings were averaged for each of the six morphing stages. Five s were given for the selection stage. Only 2.5 s of face presentation were used for analyses. The delay between the start of the upcoming fMRI frame and the start of the trial varied randomly between intervals of 100 and 2350 ms. The self-face recognition experiment described in (Knyazev et al. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) used the same stimuli, but the participant\u0026rsquo;s task was to make a binary decision (i.e., me or not me).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3. fMRI Data acquisition and preprocessing\u003c/h2\u003e\u003cp\u003efMRI data were acquired using an EPI sequence on a Philips Ingenia 7FN8GDI 3.0 T scanner (Amsterdam, Netherlands). The first five scans at the beginning of each run were discarded to ensure the effect of scanner equilibration (TR\u0026thinsp;=\u0026thinsp;2.5 s, TE\u0026thinsp;=\u0026thinsp;35 ms, flip angle\u0026thinsp;=\u0026thinsp;90\u0026deg;, FOV\u0026thinsp;=\u0026thinsp;100, 96\u0026times;94 matrix, 25 slices of 5 mm thickness, no gap). T1-weighted high-resolution (1 mm) structural scans were obtained using a 3D MP-GR sequence (TR\u0026thinsp;=\u0026thinsp;7.8 ms, TE\u0026thinsp;=\u0026thinsp;3.76 ms, matrix 252\u0026times;227). Preprocessing was performed using the SPM-12 toolbox and included slice-time correction, realignment using rigid body transformation, coregistration, normalization to the Montreal Neurological Institute (MNI) template, and resampling to 2\u0026times;2\u0026times;2 mm resolution.\u003c/p\u003e\u003cp\u003eBecause MVPA methods are sensitive to signal-to-noise ratio (Dimsdale-Zucker and Ranganath \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2018\u003c/span\u003e), we additionally denoised the data using the GLMsingle software that enables accurate estimation of single-trial fMRI responses (Prince et al. \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). A GLM design matrix with separate regressors for each trial presentation and six realignment parameters as additional regressors of no interest was constructed, and for each voxel, a custom hemodynamic response function was identified from a library of candidate functions. Cross-validation was used to derive a set of noise regressors from voxels unrelated to the experiment, and finally the betas were regularized on a voxel-wise basis using ridge regression (Prince et al. \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) and smoothed (full-width half-maximum 4 mm).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e2.4. GLM analysis\u003c/h2\u003e\u003cp\u003eUnivariate GLM analysis was conducted using the SPM-12 toolbox. Denoised beta weights were averaged across trials and runs for each participant separately, further smoothed (full-width half-maximum 8 mm), and subjected to a second-level random-effects analysis to allow for population inferences. Three-way ANOVA was used to reveal the effects of three experimental factors, including the target (participant\u0026rsquo;s own face or his\\her friend\u0026rsquo;s face), the familiarity of the opposite face (i.e., morphed toward familiar or unfamiliar person), and similarity ratings (six levels), and their interactions.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.5. Deriving the signature\u003c/h2\u003e\u003cp\u003eTo create the signatures of self-face similarity (SFSS) and friend-face similarity (FFSS), we used the Least Absolute Shrinkage and Selection Operator-Regularized Principal Components Regression (LASSO-PCR; Wager et al. \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2011\u003c/span\u003e), as implemented in CanlabCore Tools v2 (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/canlab/CanlabCore\u003c/span\u003e\u003cspan address=\"https://github.com/canlab/CanlabCore\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, accessed on 30 December 2024). This is a regression-based method, which is frequently used for predicting continuous variables, such as e.g. self-reported ratings of pain intensity (e.g., Wager et al. \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2013\u003c/span\u003e). To derive the SFSS and FFSS across contexts, we aggregated the experimental conditions with familiar and unfamiliar opposite faces. In additional analyses the signatures were generated for each condition separately. In each participant, in each run, and for each target (i.e., me and friend) separately, the beta images obtained in single trials were averaged across the six morphing stages and used as input to the LASSO-PCR algorithm to predict participants\u0026rsquo; ratings, which also were averaged across the six morphing stages. Additional analyses were conducted in which participants\u0026rsquo; ratings were averaged into five categories regardless of morphing stages, and the corresponding beta weights were averaged across these categories. Some participants did not use the full range of allowed ratings and these participants were excluded from these analyses. This left 40 participants in the \u0026lsquo;friend\u0026rsquo; condition and 42 participants in the \u0026lsquo;me\u0026rsquo; condition. Comparison of the results did not reveal a significant difference, so we report the results obtained on the full dataset. We employed a 5-fold cross-validation, in which all participants were randomly assigned to 5 subsamples of 10 or 11 participants and the model is fitted on 4 of the subsamples and tested on the left-out subsample. This is repeated 5 times with each subsample being the testing set once to assess cross-validation performance (Wager et al. \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2011\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eJabakhanji et al. (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) argue that machine-learning-derived fixed-weights decoders are not required to successfully predict a behavioral outcome, which can be achieved based on the information present in the initial beta weight map used to train the decoders (i.e., encoders). To test this, we compared predictive performance of SFSS with that of the initial beta-map and the contrast-based (i.e., me vs not me) beta-map.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003e2.6. Searchlight regression analysis\u003c/h2\u003e\u003cp\u003eSearchlight analysis was performed using CoSMo MVPA toolbox (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.cosmomvpa.org/\u003c/span\u003e\u003cspan address=\"http://www.cosmomvpa.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) (Oosterhof et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). For each voxel in the brain, a spherical neighborhood (SN) comprising the 200 voxels closest to the central voxel in the brain grey matter mask was defined and the LASSO-PCR algorithm described above and a 5-fold cross-validation scheme were used to predict participants\u0026rsquo; scores based on each SN. Prediction accuracy for each participant was defined as the correlation between that participant\u0026rsquo;s ratings and the responses calculated by the LASSO-PCR algorithm for that participant. The significance of these correlations was tested using a one-tailed permutation test (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001, 20,000 permutations) (Crosse et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). For SNs that passed the test, the overall prediction-outcome correlation was summarized at its center voxel.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003e2.7. The whole brain vs. ROI analysis\u003c/h2\u003e\u003cp\u003eThe above-described procedure was performed using as input the beta images covering either the whole brain (limited to the gray-matter mask without the brain stem and cerebellum) or selected ROIs. We selected six large ROIs, each of which included a set of brain regions involved in functions related to face recognition, self-processing, or other putatively relevant processes. These regions were identified based on the fMRI meta-analytic resource Neurosynth (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://identifiers.org/neurovault.collection:2099\u003c/span\u003e\u003cspan address=\"https://identifiers.org/neurovault.collection:2099\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, Yarkoni et al. \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2011\u003c/span\u003e) and the search terms \u0026lsquo;early visual\u0026rsquo; (136 studies, 949 voxels), \u0026lsquo;inferior occipital\u0026rsquo; (116 studies, 690 voxels), \u0026lsquo;fusiform face\u0026rsquo; (143 studies, 1920 voxels), \u0026lsquo;face recognition\u0026rsquo; (79 studies, 1095 voxels), \u0026lsquo;self-referential\u0026rsquo; (166 studies, 1075 voxels), and \u0026lsquo;medial orbitofrontal\u0026rsquo; (121 studies, 1112 voxels). Note that \u0026lsquo;early visual\u0026rsquo; ROI represented EVC, \u0026lsquo;inferior occipital\u0026rsquo; ROI represented OFA, and \u0026lsquo;fusiform face\u0026rsquo; ROI represented FFA.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003e2.8. Representational similarity analysis\u003c/h2\u003e\u003cp\u003eWe used the CoSMoMVPA toolbox to perform RSA for the five selected ROIs (Oosterhof et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). The goal of RSA was to compare the similarity of patterns of blood oxygen level dependent (BOLD) voxel-wise responses to each picture with the similarity of participants\u0026rsquo; ratings of these pictures. To construct neural representational dissimilarity matrix (RDM), we used the inverse of Spearman rank correlation coefficients (Popal et al. \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). For each subject separately, multi-voxel activation patterns from a ROI were averaged for each rating category (six categories) across runs and conditions (i.t., familiar and unfamiliar opposite face) for the \u0026lsquo;self\u0026rsquo; and \u0026lsquo;friend\u0026rsquo; targets separately and used to calculate inverse correlations between these averages. Behavioral RDMs were calculated as the absolute difference between participant\u0026rsquo;s similarity scores averaged respectively. These behavioral RDMs were used as the reference RDM, and the relatedness of each ROI RDM to the reference RDM was calculated for each subject using Spearman rank correlations between the upper triangles of the respective RDMs (Nili et al. \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). In all cases the resulting subject-specific Fischer-z-transformed correlation coefficients were entered into a one-sample permutation t-test (20,000 iterations) against the null hypothesis of a mean of zero (Crosse et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003e2.9. Mediation analysis\u003c/h2\u003e\u003cp\u003eAccording to the formal mediation model outlined in Baron and Kenny (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e1986\u003c/span\u003e), we need to have significant relationships between the independent variable (here, the morphing stages), the dependent variable (here, the facial similarity rating), and the mediator (here, the signature response). In addition, the relationship between the independent and dependent variables should decrease or become not significant after controlling for the mediator effect. To test the mediation model, we used the mediation function from the M3 Mediation Toolbox (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/canlab/MediationToolbox\u003c/span\u003e\u003cspan address=\"https://github.com/canlab/MediationToolbox\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, Wager et al. \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). For this analysis, we combined the familiar and unfamiliar opposite face conditions, thus obtaining 108 trials for each participant. The mediation path model was fitted in each participant and, at the second level, the significance of effects in the group of participants was tested using bias-corrected accelerated bootstrap tests (10,000 samples) (Shrout and Bolger \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2002\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003e2.10. Statistical testing\u003c/h2\u003e\u003cp\u003eIn the traditional GLM analysis, the main effect of each factor, as well as their interactions were estimated at a voxel-level threshold of p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 uncorrected for multiple comparisons and a cluster-level threshold of p\u0026thinsp;\u0026lt;\u0026thinsp;0.05, FWE-corrected.\u003c/p\u003e\u003cp\u003eAfter deriving the signature, for display only, we thresholded the signature map using a 10000-sample bootstrap procedure at \u003cem\u003eq\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05, FDR corrected (Zhou et al. \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Statistical maps were visualized and cluster information (including location of activation peaks, the size of each cluster, and the closest Talairach label for each peak coordinate) was extracted using NeuroElf software (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://neuroelf.net\u003c/span\u003e\u003cspan address=\"http://neuroelf.net\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eHaufe et al. (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2014\u003c/span\u003e) using simulated data showed that high weight ascribed by a classification algorithm to a particular brain region does not necessarily mean that this region is directly involved in the cognitive process of interest. Therefore, a more accurate interpretation of the predictive brain regions could be achieved by comparing the multivariate patterns obtained by means of machine learning methods with \u0026lsquo;activation patterns\u0026rsquo; showing the direction of the relationship between each brain region activity and the cognitive process of interest without controlling for other variables. In statistical terms, these activation patterns are similar to the so-called structure coefficients, whose usefulness in interpreting multiple regression results has been recognized (Courville and Thompson \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2001\u003c/span\u003e). According to Haufe et al. (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2014\u003c/span\u003e), voxels showing significant predictive weights and structure coefficients are important because they are both directly correlated with outcomes and are predictive after accounting for other brain regions. For computing within-participant activation patterns we used the formula presented in Zhou et al. (\u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), implemented in CanlabTools\u0026rsquo; \u003cem\u003efast_haufe\u003c/em\u003e function. The significant brain regions were determined using a one-sample t test thresholded at q\u0026thinsp;\u0026lt;\u0026thinsp;0.05, FDR corrected.\u003c/p\u003e\u003cp\u003eTo test the predictive performance of signatures developed based on whole brain volume or selected ROIs, signature responses were calculated as dot product between the signature weights and the brain image of estimated activity (i.e., beta weights, Han et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). These responses were then correlated with behavioral data (e.g., similarity scores). In addition, the predictive performance of signatures was evaluated by prediction-outcome correlations based on the data obtained in the process of cross-validation. The significance of within-person correlations was tested using a one-sample permutation test (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001, 20,000 permutations) (Crosse et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), which also was used for statistical comparison of Fischer-z-transformed correlation coefficients.\u003c/p\u003e\u003cp\u003eShort-term test-retest reliability of the mean signature response was determined by the intra-class correlation coefficient (ICC; Koo and Li \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Han et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Reliability was assessed for all participants in three runs within one experiment. We used a two-way mixed-effects model with time as a fixed effect and participants as a random effect, as implemented in SPSS package.\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003e3.1. Behavioral data\u003c/h2\u003e\u003cp\u003eA repeated-measures ANOVA with the factors of target (me vs friend), familiarity (familiar vs unfamiliar opposite face) and similarity (six levels), and similarity scores as dependent variable showed significant main effects of familiarity (F\u003csub\u003e1,53\u003c/sub\u003e = 29.88, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, \u003cem\u003eη\u003c/em\u003e\u003csub\u003e\u003cem\u003ep\u003c/em\u003e\u003c/sub\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.361) and similarity (F\u003csub\u003e5,265\u003c/sub\u003e = 1033.01, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, \u003cem\u003eη\u003c/em\u003e\u003csub\u003e\u003cem\u003ep\u003c/em\u003e\u003c/sub\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.951), as well as an interaction between similarity and familiarity (F\u003csub\u003e5,265\u003c/sub\u003e = 19.83, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, \u003cem\u003eη\u003c/em\u003e\u003csub\u003e\u003cem\u003ep\u003c/em\u003e\u003c/sub\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.272). As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, for both targets, similarity scores are lower when an unfamiliar opposite face is used for morphing. Similarity scores strongly correlated with morphing stages (mean r(54)\u0026thinsp;=\u0026thinsp;0.94, SD\u0026thinsp;=\u0026thinsp;0.04). Paired samples permutation t-tests on Fischer-z-transformed correlation coefficients found no difference between different combinations of target and familiarity conditions (all ps\u0026thinsp;\u0026gt;\u0026thinsp;0.12).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003e3.2. Univariate GLM analysis\u003c/h2\u003e\u003cp\u003eThree-way ANOVA was performed on denoised beta weights with the factors of target (\u0026lsquo;seld\u0026rsquo; vs \u0026lsquo;friend\u0026rsquo;), familiarity (familiar vs unfamiliar opposite face) and similarity (six levels). The main effect of target did not produce significant effects. The main effect of familiarity was significant in the left middle temporal gyrus (T\u003csub\u003e1,1272\u003c/sub\u003e = 4.75; x = -45, y = -61, z\u0026thinsp;=\u0026thinsp;17; K\u003csub\u003eE\u003c/sub\u003e = 218; P\u003csub\u003eFWE\u0026minus;corr\u003c/sub\u003e = 0.001). The main effect of similarity was significant in a cluster spanning many brain regions and centered in the right lingual gyrus (F\u003csub\u003e5,1272\u003c/sub\u003e = 196.39; x\u0026thinsp;=\u0026thinsp;12, y = -85, z = -7; K\u003csub\u003eE\u003c/sub\u003e = 20147; P\u003csub\u003eFWE\u0026minus;corr\u003c/sub\u003e \u0026lt; 0.001). Other effects were not significant. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e shows the localization of the main effect of similarity, as well as contrast estimates and 90% confidence intervals of the left and right lingual gyrus response at each similarity rating point. It can be seen that the right lingual gyrus response shows a gradual increase in activation with increasing similarity ratings, while the left lingual gyrus response shows the opposite dynamics.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003e3.3. Deriving the self-face (SFSS) and friend-face (FFSS) similarity signatures\u003c/h2\u003e\u003cp\u003eFigures \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB show SFSS and FFSS thresholded at q\u0026thinsp;\u0026lt;\u0026thinsp;0.05, FDR corrected. Overall prediction-outcome correlation based on the data obtained in the process of cross-validation was r(1944)\u0026thinsp;=\u0026thinsp;0.74 for SFSS and r(1944)\u0026thinsp;=\u0026thinsp;0.72 for FFSS. Mean within-person correlation was r(36)\u0026thinsp;=\u0026thinsp;0.73\u0026thinsp;\u0026plusmn;\u0026thinsp;0.15, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 for SFSS and r(36)\u0026thinsp;=\u0026thinsp;0.72\u0026thinsp;\u0026plusmn;\u0026thinsp;0.11, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 for FFSS. In both cases, signature responses showed a perfect (r\u0026thinsp;=\u0026thinsp;1) correlation with similarity scores obtained in the same condition. SFSS responses also significantly correlated with similarity scores in the \u0026lsquo;friend\u0026rsquo; condition (r(1944)\u0026thinsp;=\u0026thinsp;0.66, t\u0026thinsp;=\u0026thinsp;32.5, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and FFSS responses significantly correlated with similarity scores in the \u0026lsquo;self\u0026rsquo; condition (r(1944)\u0026thinsp;=\u0026thinsp;0.54, t\u0026thinsp;=\u0026thinsp;19.6, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eA comparison of Figs.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB shows that the distributions of most predictive weights look similar, however, the overall correlation between SFSS and FFSS patterns was only r(47158)\u0026thinsp;=\u0026thinsp;0.175, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001. Both patterns showed comparable correlations with the distribution of F values obtained in the univariate GLM analysis of the main effect of similarity and presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA \u0026ndash; r(47158)\u0026thinsp;=\u0026thinsp;0.127, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 and r(47158)\u0026thinsp;=\u0026thinsp;0.124, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 for SFSS and FFSS, respectively. Figures\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD show the \u0026lsquo;activation patterns\u0026rsquo; (i.e., structure coefficients) averaged across participants and FDR-thresholded at q\u0026thinsp;\u0026lt;\u0026thinsp;0.05. It appears that most regions with significant model weights also show significant activation. Spatial correlations between signatures and activation patterns were r(47158)\u0026thinsp;=\u0026thinsp;0.107, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 and r(47158)\u0026thinsp;=\u0026thinsp;0.355, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 for SFSS and FFSS, respectively. The overlap between the two signatures and corresponding activation patterns is shown in Figs.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eE and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eF. In both cases most prominent clusters included the right and left lingual gyrus.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003e3.4. Exploring the properties of SFSS\u003c/h2\u003e\u003cdiv id=\"Sec18\" class=\"Section3\"\u003e\u003ch2\u003e3.4.1. Predictive performance of encoders\u003c/h2\u003e\u003cp\u003eJabakhanji et al. (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) argue that multivariate fixed weights decoders perform no better than the GLM-derived encoders. To test this, we assessed the predictive performance of the GLM-derived map of the main effect of similarity. When used as a decoder, this map\u0026rsquo;s expression showed a correlation of 0.22 (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) with similarity scores in the \u0026lsquo;self\u0026rsquo; condition. For within-person prediction, mean correlation was 0.25 (t(54)\u0026thinsp;=\u0026thinsp;8.5, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec19\" class=\"Section3\"\u003e\u003ch2\u003e3.4.2. Spatial smoothing of voxel weights\u003c/h2\u003e\u003cp\u003eTo investigate the effect of spatial smoothing of the SFSS weights on its performance, we used a Gaussian filter with increasing width, from 1 mm to 20 mm. At 1 mm smoothing, the performance deterioration was very small (r(1944)\u0026thinsp;=\u0026thinsp;0.9948), but the difference was already significant, as shown by a permutation paired samples t-test for all participants (t\u003csub\u003e54\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;9.12, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Performance deteriorated further with increasing filter width, reaching a prediction-outcome correlation of 0.71 at 20 mm smoothing. We also constructed a \u0026lsquo;sign\u0026rsquo; version of SFSS in which all positive voxels were assigned a value of +\u0026thinsp;1 and negative voxels were assigned a value of -1. The performance of this version was significantly worse than the original SFSS (r\u0026thinsp;=\u0026thinsp;0.57).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec20\" class=\"Section3\"\u003e\u003ch2\u003e3.4.3. Effect of the number of features/voxels\u003c/h2\u003e\u003cp\u003eNext, we analyzed the effect of the number of voxels randomly selected in the whole brain on prediction success. An increasing number of voxels (starting at 100 and in increments of 100) were repeatedly (100 iterations for each step) randomly selected from the whole brain and used to predict the similarity scores in the \u0026lsquo;self\u0026rsquo; condition. As can be seen in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, the model performance is unstable below 2500 voxels, then increases asymptotically and reaches a plateau, approaching (but not reaching) the performance of a signature based on whole brain analysis when a random sample of about 5000 voxels is selected. Then, following the suggestion of Zhou et al. (\u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), 10,000 voxels were randomly selected in the brain. The difference between averaged over 1000 iterations performance of 10,000 randomly selected voxels and all voxels in the brain was still significant across participants, as shown by the paired samples permutation test (t\u003csub\u003e54\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;4.2; p\u0026thinsp;=\u0026thinsp;0.0001). Further increase of the number of voxels showed that the difference remained to be significant up to 25,000 voxels (p\u0026thinsp;=\u0026thinsp;0.020), but was not significant with 30,000 voxels (p\u0026thinsp;=\u0026thinsp;0.477).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec21\" class=\"Section2\"\u003e\u003ch2\u003e3.5. Within- and between-person SFSS reliability\u003c/h2\u003e\u003cp\u003eThe overall short-term (for the three runs within one experiment) test-retest reliability of SFSS response averaged across trials was excellent (ICC\u0026thinsp;=\u0026thinsp;0.97, F\u003csub\u003e53,106\u003c/sub\u003e = 33.6, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). For run 1 versus run 3 it also was excellent (ICC\u0026thinsp;=\u0026thinsp;0.94, F\u003csub\u003e53,53\u003c/sub\u003e = 17.4, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e\u003cp\u003eAn important question is to what extent a population-level signature such as the SFSS can predict each participant\u0026rsquo;s similarity score on each trial. To test this, we applied the SFSS to each participant\u0026rsquo;s single-trial beta map to obtain instantaneous signature responses in each trial. For each participant separately, Pearson\u0026rsquo;s correlation coefficient was calculated between these responses and the behavioral responses on each trial. These coefficients were Fischer-z-transformed and entered into a one-sample t-test to assess its statistical significance. Mean (SD) correlations were 0.78 (0.06), t\u003csub\u003e53\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;93.0, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, Cohen\u0026rsquo;s d\u0026thinsp;=\u0026thinsp;13.0. In all participants, correlations were significant at q\u0026thinsp;\u0026lt;\u0026thinsp;0.05, FDR-corrected level. We also tested the performance of SFSS using pooled single-trial data for all participants. In this case, the correlation was 0.79 (0.77\u0026ndash;0.80), Cohen\u0026rsquo;s d\u0026thinsp;=\u0026thinsp;2.58. Another question is how well the SFSS can predict individual differences in the ability to assess facial similarity. To answer this question, we assessed each participant\u0026rsquo;s sensitivity to the similarity of the images to their own face as a correlation between the morphing stages and similarity ratings. The same indices were calculated for SFSS responses. The Pearson correlation coefficient for behavioral and neural indices of sensitivity was r(54)\u0026thinsp;=\u0026thinsp;0.33, p\u0026thinsp;=\u0026thinsp;0.017, Cohen\u0026rsquo;s d\u0026thinsp;=\u0026thinsp;0.70.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec22\" class=\"Section2\"\u003e\u003ch2\u003e3.6. Mediation analysis\u003c/h2\u003e\u003cp\u003eThe results reported so far show that facial similarity ratings correlate with the degree of image morphing. SFSS response also correlated with the degree of image morphing r(1944)\u0026thinsp;=\u0026thinsp;0.88. In addition, SFSS response correlates with similarity ratings, as shown by the prediction-outcome correlation described above. Thus, the initial conditions for testing of the mediation model are fulfilled. The mediation analysis showed a significant effect of mediation (B\u0026thinsp;=\u0026thinsp;2.01, STE\u0026thinsp;=\u0026thinsp;0.06, t\u003csub\u003e53\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;36.26; p\u0026thinsp;=\u0026thinsp;0.0004). The strength of association between the morphing stage and the facial similarity ratings (B\u0026thinsp;=\u0026thinsp;2.01, STE\u0026thinsp;=\u0026thinsp;0.06, t\u003csub\u003e53\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;36.32; p\u0026thinsp;=\u0026thinsp;0.0004) became no significant (B\u0026thinsp;=\u0026thinsp;0.00, STE\u0026thinsp;=\u0026thinsp;0.00, t\u003csub\u003e53\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.56; p\u0026thinsp;=\u0026thinsp;0.579) when controlling for the effect of the signature response. Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e shows results of bootstrap tests of the mediation model path coefficients. It can be seen that the significant slope \u003cb\u003ec\u003c/b\u003e (degree of morphing\u0026rarr;similarity scores without controlling for signature response) becomes no significant (\u003cb\u003ec\u0026rsquo;\u003c/b\u003e) while controlling for signature response.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec23\" class=\"Section2\"\u003e\u003ch2\u003e3.7. Searchlight analysis\u003c/h2\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e shows the results of the searchlight regression analysis. The highest accuracy scores were found in the right and left lingual gyrus.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec24\" class=\"Section2\"\u003e\u003ch2\u003e3.8. ROI-based signatures\u003c/h2\u003e\u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e shows prediction-outcome correlations obtained for each ROI in the process of cross-validation in the \u0026lsquo;self\u0026rsquo; and \u0026lsquo;friend\u0026rsquo; conditions. The highest correlations were observed for the EVC and OFA, while the FFA showed the smallest correlation. The correlation between the size of the ROI and its predictive performance was not significant (p\u0026thinsp;\u0026gt;\u0026thinsp;0.4).\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePrediction-outcome correlations for ROI-based signatures in the \u0026lsquo;self\u0026rsquo; and \u0026lsquo;friend\u0026rsquo; conditions.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eROI\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u0026lsquo;self\u0026rsquo;\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u0026lsquo;friend\u0026rsquo;\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEarly visual\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.69\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.66\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eInferior occipital\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.34\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.32\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFusiform face area\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.04\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.05\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFace recognition\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.18\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.16\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSelf-referential\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.22\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.20\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMedial orbitofrontal\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.06\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.09\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"3\"\u003eAll correlations are significant at p\u0026thinsp;\u0026lt;\u0026thinsp;0.05, FDR-corrected.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec25\" class=\"Section2\"\u003e\u003ch2\u003e3.9. Representational similarity analysis\u003c/h2\u003e\u003cp\u003eSpearman correlations between ROI RDMs and behavioral RDMs averaged across participants in the \u0026lsquo;self\u0026rsquo; and \u0026lsquo;friend\u0026rsquo; conditions are presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Significance of correlations was tested using a one-sample permutation t-test across participants. The highest correlation in both conditions could be noted for EVC. For self-referential and MOC ROIs, correlations are smaller in the \u0026lsquo;friend\u0026rsquo; than in the \u0026lsquo;self\u0026rsquo; condition. For MOC the correlation was significant in the \u0026lsquo;self\u0026rsquo;, but was not significant in the \u0026lsquo;friend\u0026rsquo; condition. For self-referential ROI it was significant in both cases, but a paired-samples permutation t-test showed that it was significantly higher in the first case (t\u003csub\u003e52\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;2.01, p\u0026thinsp;=\u0026thinsp;0.023).\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eRSA results. Spearman correlations between ROI RDMs and behavioral RDMs averaged across participants. Significance of correlations was tested using a one-sample permutation t-test across participants.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eROI\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u0026lsquo;self\u0026rsquo;\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u0026lsquo;friend\u0026rsquo;\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEarly visual\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.64*\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.62*\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eInferior occipital\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.22*\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.23*\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFusiform face area\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.21*\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.20*\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFace recognition\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.14*\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.12*\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSelf-referential\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.25*\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.13*\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMedial orbitofrontal\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.08*\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"3\"\u003e* correlation is significant at p\u0026thinsp;\u0026lt;\u0026thinsp;0.05, FDR-corrected.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec26\" class=\"Section2\"\u003e\u003ch2\u003e3.10. Decomposing SFSS and FFSS into common and distinct components\u003c/h2\u003e\u003cp\u003eAs we discussed in the Introduction, the mental processes whose signatures are SFSS and FFSS must have both common and distinct components - the former are related to the evaluation of similarity of faces regardless of the target with which the similarity is evaluated, and the latter related specifically to the target of assessment. To decompose the SFSS and FFSS into these components, we used principal component analysis (PCA). The first component, which presumably accounted for the common features of SFSS and FFSS, explained 59% of the variance. Using the \u003cem\u003epcares\u003c/em\u003e Matlab function we created signatures of both reconstructed (REC) and residual (RES) patterns after extracting this component. Thresholded maps of these patterns are shown in Figs.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eA and \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eB. It can be seen that the REC map has strong positive values in the right lingual gyrus and strong negative values in the left lingual gyrus. The RES pattern has most prominent positive and negative values in occipital, temporal, and orbitofrontal regions. Responses of REC and RES patterns were then calculated in the \u0026lsquo;self\u0026rsquo; and \u0026lsquo;friend\u0026rsquo; conditions and correlated with similarity scores. Correlations between REC responses and similarity scores were 0.88 and 0.92 in the \u0026lsquo;self\u0026rsquo; and \u0026lsquo;friend\u0026rsquo; conditions, respectively. For RES they were \u0026minus;\u0026thinsp;0.52 and 0.55, respectively. This confirms that REC reflects features common to both conditions, while RES reflects features that distinguish them. Next, we regressed out REC from beta weights in original data and trained the LASSO-PCR algorithm with 5-fold cross-validation to predict similarity scores in the \u0026lsquo;self\u0026rsquo; and \u0026lsquo;friend\u0026rsquo; conditions. The obtained purified self-face similarity (PSFSS) signature looked similar to RES. Correlations between PSFSS responses and similarity scores were 1.0 and \u0026minus;\u0026thinsp;0.09 in the \u0026lsquo;self\u0026rsquo; and \u0026lsquo;friend\u0026rsquo; conditions, respectively. Correlations between purified friend-face similarity signature (PFFSS) responses and similarity scores were \u0026minus;\u0026thinsp;0.07 and 0.99 in the \u0026lsquo;self\u0026rsquo; and \u0026lsquo;friend\u0026rsquo; conditions, respectively. Thus, both PSFSS and PFFSS predict similarity scores in their own conditions no worse than the original SFSS and FFSS, but do not predict similarity scores in the opposite condition. PSFSS and PFFSS patterns correlated negatively with each other (r(47158) = -0.76). We then analyzed the overlap between the positive and negative weights of PSFSS and PFFSS. They were transformed into z-scores and thresholded at z\u0026thinsp;=\u0026thinsp;1.96. Next, overlaps were analyzed for positive vs positive, negative vs negative, and positive vs negative weights separately. There were no overlaps for positive vs positive and negative vs negative weights. Only overlaps for positive vs negative weights were found in the fusiform and orbital gyri (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eC).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec27\" class=\"Section2\"\u003e\u003ch2\u003e3.10. Overlap with the self-face-recognition signature (SFRS)\u003c/h2\u003e\u003cp\u003eThe SFRS was developed previously using the same analytic approach and experimental paradigm, which was similar to the paradigm of the present study in all aspects except one: on each trial, participants were required to make a forced choice decision (\u0026lsquo;me\u0026rsquo; or \u0026lsquo;not me\u0026rsquo;). We first tested whether the SFSS response could predict self-recognition in a forced choice task, and conversely, whether the SFRS response could predict a face\u0026rsquo;s similarity to one\u0026rsquo;s own face score in this study task. A one-sample permutation t-test across participants was used to determine the significance of the correlation between prediction and outcome in each participant. It showed a significant prediction-outcome correlation in both cases (mean r\u0026thinsp;=\u0026thinsp;0.09, t\u003csub\u003e53\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;2.11, p\u0026thinsp;=\u0026thinsp;0.041 and mean r\u0026thinsp;=\u0026thinsp;0.17, t\u003csub\u003e53\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;5.98, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 for SFSS and SFRS, respectively). When FFSS and, respectively, friend\u0026rsquo;s face similarity data were used instead of SFSS and self-face similarity data, the prediction-outcome correlations were not significant (mean r\u0026thinsp;=\u0026thinsp;0.006, t\u003csub\u003e53\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.16, p\u0026thinsp;=\u0026thinsp;0.874 and mean r\u0026thinsp;=\u0026thinsp;0.04, t\u003csub\u003e53\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;1.57, p\u0026thinsp;=\u0026thinsp;0.12 for FFSS and SFRS, respectively).\u003c/p\u003e\u003cp\u003eTo identify commonalities in SFSS and SFRS, we again used PCA. The first component, which accounted for the common features of SFSS and SFRS, explained 98% of the variance. Using the \u003cem\u003epcares\u003c/em\u003e Matlab function we reconstructed the signature of this component. This signature response showed significant overall correlations with both self-recognition in a forced choice task, r(704)\u0026thinsp;=\u0026thinsp;0.82, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, and self-face similarity score in this study task, r(1944)\u0026thinsp;=\u0026thinsp;0.71, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001. A thresholded map of this signature is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eD. It has positive and negative weights in the right fusiform gyrus and right and left orbitofrontal regions. It also showed positive weights in the right lingual gyrus and negative weights in the left lingual gyrus.\u003c/p\u003e\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eIn this study, we aimed to explore the brain signature of the assessment of facial similarity to one\u0026rsquo;s own face (SFSS). We aimed to compare the results of a traditional mass-univariate data analysis approach with those of MVPA and to examine the properties of the signature obtained using MVPA methods. Besides, we hypothesized that this signature may have two components: one related to the assessment of face similarity independently of the object with which the similarity is assessed, and the other related specifically to the assessment of face similarity to one\u0026rsquo;s own face. To distinguish between these two components, we included a task in which participants rated the similarity of morphed images to their friend\u0026rsquo;s face. In addition, we compared the SFSS to a self-recognition signature previously derived in an experimental paradigm that does not require explicit face similarity assessment. We expected that face similarity assessment in general should be reflected predominantly in right-sided changes in EVC and FSBA, whereas similarity assessment specifically to one\u0026rsquo;s own face should engage brain regions associated with self-referential processing.\u003c/p\u003e\u003cp\u003eBehavioral results show that participants were able to reliably judge the actual degree of similarity of morphed images to target faces independently of target and familiarity of the opposite face, although similarity scores tended to be higher when a familiar face was used. The familiarity factor also had a significant effect in univariate GLM analysis, which was observed in the left middle temporal gyrus, whose role in assessing face familiarity, presumably related to providing access to face-related semantic knowledge, has been noted previously (Elfgren et al. \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Pourtois et al. \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2005\u003c/span\u003e). In general, however, GLM analysis results indicate that the observed effects are preferentially associated with the assessment of face similarity regardless of the target with which the similarity is assessed. This is evident from the fact that the main effect of target was not significant, whereas the main effect of similarity spanned multiple brain regions, reaching a maximum in the right EVC. The right-sided dominance of the similarity effect is consistent with the known dominance of the right hemisphere in the perception of faces (\u0026Aring;sberg Johnels et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Harrison and Strother \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Hougaard et al. \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Levine et al. \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e1988\u003c/span\u003e). Notably, right-sided EVC activation gradually increases with increasing similarity ratings, while left-sided EVC activation exhibits the opposite dynamics (see Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB). This shows that the initial information for face similarity assessment is already present in EVC (see e.g., Lyons and Morikawa \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Yue et al. \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2012\u003c/span\u003e), which then passes to higher levels of face similarity processing (Duchaine and Yovel 2016).\u003c/p\u003e\u003cp\u003eComparison of the results of univariate and multivariate analyses shows that the prognostically most significant SFSS weights coincide with the major loci identified in the GLM analysis. This mostly relates to the right and left EVC areas. Notably, the right-sided EVC contains positive SFSS weights and the left-sided EVC contains negative weights, which is consistent with the dynamics identified in the GLM analysis. Comparing the SFSS weights to the thresholded \u0026lsquo;activation pattern\u0026rsquo; again shows that the right and left EVC regions are coincidence locations. These regions also showed the highest accuracy rates in the local neighborhood-based searchlight regression analysis. The RSA results and the prediction of similarity scores based on ROIs further support the idea that EVC areas are crucial for face similarity estimation. Indeed, the EVC provided an overall prediction-outcome correlation similar to that of SFSS (see Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), and showed the highest relatedness estimates in RSA (see Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Noteworthy, FFA scarcely allowed the prediction of similarity scores and showed considerably lower relatedness estimates in RSA. Thus, the main findings from different analysis methods are consistent in showing the primacy of EVC for face similarity assessment.\u003c/p\u003e\u003cp\u003eThe overall prediction-outcome correlation obtained for SFSS is comparable to that obtained in other relevant studies (e.g., Knyazev et al. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Zhou et al. \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), and SFSS response showed a perfect correlation with behavioral data and excellent short-term test-retest reliability. Moreover, it showed a large effect size in predicting the instantaneous (i.e., in each trial) similarity score for each participant and was significantly associated with individual differences in sensitivity to the similarity of the images to participant\u0026rsquo;s own face. Most impressively, the mediation analysis showed that the SFSS response acts as a mediator between the degree of similarity objectively present in the image and the subjective assessment of similarity. All this suggests that the SFSS is a robust brain signature of the mental processes involved in assessing the similarity of face images to one\u0026rsquo;s own face.\u003c/p\u003e\u003cp\u003eFollowing Jabakhanji et al.\u0026rsquo;s (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) suggestion, we explored the properties of this signature. Our results contradict the claim that predictive performance of machine-learning-based decoders is always no better than that of the mass-univariate encoders used for their derivation. In our case, the predictive performance of the GLM-derived map of the main effect of similarity was significantly worse than that of SFSS. Spatial smoothing of the SFSS weights degraded its predictive performance as early as 1 mm smoothing, and the \u0026lsquo;sign\u0026rsquo; version of the SFSS performed significantly worse than the original SFSS. Analyzing the effect of the number of randomly selected voxels showed that more than a half of voxels were needed to achieve the prediction success rate obtained using the whole brain. These properties of SFSS differ from those of SFRS described earlier (Knyazev et al. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). It can be hypothesized that they depend on the characteristics of the mental processes that are being decoded. In particular, self-face similarity assessment is a more difficult task than self-face recognition, and, respectively, the information needed to decode it must be more specific and structured.\u003c/p\u003e\u003cp\u003eAfter confirming that the SFSS is a reliable brain signature for assessing similarity between a face and one\u0026rsquo;s own face, a legitimate question arises as to what mental processes it reflects. As we discussed in the Introduction, the mental processes involved in judging the similarity of images to one\u0026rsquo;s own face must have at least two components \u0026ndash; one related to judging the similarity of faces independently of the object with which the similarity is judged, and one related specifically to judging the similarity of images to one\u0026rsquo;s own face. To separate these processes, we compared the topography of the SFSS weights with that of the FFSS and SFRS, the former of which being associated with assessing the similarity of faces other than one\u0026rsquo;s own, and the latter with recognizing one\u0026rsquo;s own face without explicitly assessing the degree of similarity.\u003c/p\u003e\u003cp\u003eBoth SFSS and FFSS responses significantly predicted similarity ratings obtained not only in their own condition, but also in the opposite condition, and their topography was visually very similar. PCA reconstruction of the commonalities between SFSS and FFSS showed that they are mainly related to the right and left regions of the EVC, which, given the above discussion, are heavily involved in faces similarity assessment. Regressing these commonalities out from the beta weights in the raw data yielded signatures that selectively predicted similarity scores in each condition but not in the other. These signatures showed no spatial overlap between both positive vs positive and negative vs negative weights. Most interestingly, overlaps in positive vs negative weights were found in the fusiform and orbital gyri (see Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eC). It can be hypothesized that, unlike EVC, which appear to be involved in assessing the similarity of faces regardless of the target to which similarity is being assessed, these regions are involved in target discrimination. In favor of this hypothesis is also the fact that these are the regions where the commonality between SFSS and SFRS is revealed (see Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eD). Thus, the right fusiform gyrus and the left OFC are the places where SFSS is similar with SFRS and is dissimilar with FFSS.\u003c/p\u003e\u003cp\u003eFrom the data presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, it can be seen that both FFA and medial OFC (mOFC) are barely predictive of similarity scores. Perhaps this is not surprising in the case of mOFC, but it is puzzling that FFA, the core area of FSBA, is so weakly associated with subjective assessments of facial similarity. These results are consistent with the view that face similarity is first processed in EVC before being transferred to other areas of FSBA (Duchaine and Yovel 2016). As for FFA, it is involved in processing other aspects of the presented faces, such as whether to pay attention to them or ignore them (e.g., Gentile and Jansma \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2010\u003c/span\u003e), or, in our case, whether it is my own or someone else\u0026rsquo;s face.\u003c/p\u003e\u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, compared to the other ROIs, the mOFC shows the least representative similarity to the behavioral data. Clearly, the mOFC is not a brain region specifically involved in face similarity assessment. Given the existing evidence of mOFC involvement in reward processing (Diekhof et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Howard and Kahnt \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) and frequently observed association of self-referential processing with activation of reward circuits (Chakraborty and Chakrabarti \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Devue et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Ota and Nakano \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Platek et al. \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Zhan et al. \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2016\u003c/span\u003e), one may speculate that self-recognition and self-face similarity assessment are more associated with reward (and, correspondingly, mOFC activation) than friend\u0026rsquo;s face similarity assessment. This partly is confirmed by the fact that mOFC ROI\u0026rsquo;s RDM was significantly associated with behavioral RDM in the \u0026lsquo;self\u0026rsquo;, but not in the \u0026lsquo;friend\u0026rsquo; condition (see Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Additionally, representation similarity of self-referential ROI was significantly higher in the \u0026lsquo;self\u0026rsquo; than in the \u0026lsquo;friend\u0026rsquo; condition, as shown by a paired samples permutation t-test across participants (t\u003csub\u003e52\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;2.05, p\u0026thinsp;=\u0026thinsp;0.046).\u003c/p\u003e\u003cp\u003eA possible limitation of this study is the relatively small sample size. However, a post hoc power calculation shows that with the effect sizes found in this study (Cohen\u0026rsquo;s d ranging from 2.58 to 13) a sample size of 54 participants achieves a power of 0.99. This is consistent with other MVPA studies generally showing high effect sizes for associations between brain signatures and mental states (e.g., Han et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Kragel et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eTo summarize, in this study, using MVPA approaches, we developed a brain signature of face similarity assessment and demonstrated its reliability in both intrapersonal and interpersonal domains in predicting subjective face similarity ratings. This signature response showed a perfect correlation with behavioral data and a large effect size in predicting the instantaneous (i.e., in each trial) similarity score for each participant. Moreover, it acted as a mediator between the degree of similarity objectively present in the image and the subjective assessment of similarity. Different analytical approaches including the traditional GLM, MVPA classification based on the whole brain, selected ROIs, and local neighborhood, and RSA converged in showing the primacy of early visual cortical areas (in particular, the right EVC) for face similarity assessment. Principal component analyses of three distinct signatures (SFSS, FFSS, and SFRS) allowed to separate brain regions associated with the assessment of face similarity independently of the target with which the similarity is judged and specifically associated with judging the similarity to one\u0026rsquo;s own face. It appears that the EVC is particularly involved in the first case whereas the fusiform gyrus and the OFC are the places where the distinction of targets is processed. The results of this analysis show the fruitfulness of considering brain signatures as representations of complex psychological processes, which can be analyzed by decomposing the signatures into their components and combining them in various ways to represent different combinations of relevant psychological states. This approach paves the way to what is considered the fundamental goal of cognitive neuroscience: establishing a correspondence between mind and brain (Kragel et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by budgetary funding for basic scientific research (theme No. 122042700001-9).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCRediT authorship contribution statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eG.G. Knyazev: Methodology, Conceptualization, Data analysis, Writing \u0026ndash; original draft. A.N. Savostyanov: Data curation, Recruiting participants, Designing experimental setup, Writing \u0026ndash; review and editing. A.V. Bocharov: Data curation, Writing \u0026ndash; review and editing. A.E. Saprigyn: Data curation, Software, Designing experimental setup, Writing \u0026ndash; review and editing. E.A. Levin: Data curation, Software, Writing \u0026ndash; review and editing. D. A. Lebedkin: Data \u0026ndash; collection and curation, Recruiting participants, Writing \u0026ndash; review and editing.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeclaration of interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNone of the authors have a conflict of interest to disclose\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data used to create the SFRS and the SFRS itself are available at https://doi.org/10.17605/OSF.IO/28HKV.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was supported by budgetary funding for basic scientific research (theme No. 122042700001-9). The authors thank Dmitrienko N.V. for help with data collection and Borisov S.V. for help with photography. MRI recordings were performed on the equipment of the International Tomography Center (ITC SB RAS).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAsakage S, Nakano T (2023) The salience network is activated during self‐recognition from both first‐person and third‐person perspectives. Hum Brain Mapp 44(2):559-570. https://doi.org/10.1002/hbm.26084\u003c/li\u003e\n\u003cli\u003e\u0026Aring;sberg Johnels J, Galazka MA, Sundqvist M, Hadjikhani N (2022) Left visual field bias during face perception aligns with individual differences in reading skills and is absent in dyslexia. Brit J Educ Psychol 00:1-10. https://doi.org/10.1111/bjep.12559\u003c/li\u003e\n\u003cli\u003eBaron RM, Kenny DA (1986) The Moderator-Mediator Variable Distinction in Social Psychological Research: Conceptual, Strategic, and Statistical Considerations. J Pers Soc Psychol 51(6):1173-1182. https://doi.org/10.1037/0022-3514.51.6.1173\u003c/li\u003e\n\u003cli\u003eChakraborty A, Chakrabarti B (2018) Looking at my own face: Visual processing strategies in self\u0026ndash;other face recognition. Front Psychol 9:121. https://doi.org/10.3389/fpsyg.2018.00121\u003c/li\u003e\n\u003cli\u003eCourville T, Thompson B (2001) Use of structure coefficients in published multiple regression articles: \u0026beta; is not enough. Educ Psychol Meas 61:229\u0026ndash;248. http://epm.sagepub.com/cgi/content/abstract/61/2/229\u003c/li\u003e\n\u003cli\u003eCrosse MJ, Foxe JJ, Molholm S (2024) PERMUTOOLS: A MATLAB Package for Multivariate Permutation Testing. arXiv preprint arXiv:2401.09401. https://github.com/mickcrosse/PERMUTOOLS\u003c/li\u003e\n\u003cli\u003eDevue C, Br\u0026eacute;dart S (2011) The neural correlates of visual self-recognition. Conscious Cogn 20(1):40-51. https://doi.org/10.1016/j.concog.2010.09.007\u003c/li\u003e\n\u003cli\u003eDevue C, Van der Stigchel S, Br\u0026eacute;dart S, Theeuwes J (2009) You do not find your own face faster; you just look at it longer. Cognition 111:114\u0026ndash;122. https://doi.org/10.1016/j.cognition.2009.01.003\u003c/li\u003e\n\u003cli\u003eDiekhof EK, Kaps L, Falkai P, Gruber O (2012) The role of the human ventral striatum and the medial orbitofrontal cortex in the representation of reward magnitude\u0026ndash;An activation likelihood estimation meta-analysis of neuroimaging studies of passive reward expectancy and outcome processing. Neuropsychologia 50(7):1252-1266. https://doi.org/10.1016/j.neuropsychologia.2012.02.007\u003c/li\u003e\n\u003cli\u003eDimsdale-Zucker HR, Ranganath C (2018) Representational similarity analyses: a practical guide for functional MRI applications. In Handbook of behavioral neuroscience, Elsevier, Vol 28, pp 509-525. https://doi.org/10.1016/B978-0-12-812028-6.00027-6\u003c/li\u003e\n\u003cli\u003eDuchaine B, Yovel G (2015) A revised neural framework for face processing. Annu Rev Vis Sci 1(1):393-416. https://doi.org/10.1146/annurev-vision-082114-035518\u003c/li\u003e\n\u003cli\u003eElfgren C, van Westen D, Passant U, Larsson EM, Mannfolk P, Fransson P (2006) fMRI activity in the medial temporal lobe during famous face processing. Neuroimage 30(2):609-616. https://doi.org/10.1016/j.neuroimage.2005.09.060\u003c/li\u003e\n\u003cli\u003eGentile F, Jansma BM (2010) Neural competition through visual similarity in face selection. Brain Res 1351:172-184. https://doi.org/10.1016/j.brainres.2010.06.050\u003c/li\u003e\n\u003cli\u003eGillihan SJ, Farah MJ (2005) Is self special? A critical review of evidence from experimental psychology and cognitive neuroscience. Psychol Bull 131(1):76\u0026ndash;97. https://doi.org/10.1037/0033-2909.131.1.76\u003c/li\u003e\n\u003cli\u003eHan X, Ashar YK, Kragel P et al (2022) Effect sizes and test-retest reliability of the fMRI-based neurologic pain signature. Neuroimage 247:118844. https://doi.org/10.1016/j.neuroimage.2021.118844\u003c/li\u003e\n\u003cli\u003eHansen F, DeBruine LM, Holzleitner IJ, Lee AJ, O\u0026apos;Shea KJ, Fasolt V (2020) Kin recognition and perceived facial similarity. J Vis 20(6):18-18. https://doi.org/10.1167/jov.20.6.18\u003c/li\u003e\n\u003cli\u003eHarrison MT, Strother L (2021) Does face-selective cortex show a left visual field bias for centrally-viewed faces?. Neuropsychologia 159:107956. https://doi.org/10.1016/j.neuropsychologia.2021.107956\u003c/li\u003e\n\u003cli\u003eHassall C, Silver A, Turk DJ, Krigolson OE (2016) We are more selfish than we think: The endowment effect and reward processing within the human medial-frontal cortex. Q J Exp Psychol 69(6):1676-1686. https://doi.org/10.1080/17470218.2015.1091849\u003c/li\u003e\n\u003cli\u003eHaufe S, Meinecke F, G\u0026ouml;rgen K, D\u0026auml;hne S, Haynes JD, Blankertz B, Bie\u0026szlig;mann F (2014) On the interpretation of weight vectors of linear models in multivariate neuroimaging. NeuroImage 87:96-110. https://doi.org/10.1016/j.neuroimage.2013.10.067\u003c/li\u003e\n\u003cli\u003eHaxby JV, Hoffman EA, Gobbini MI (2000) The distributed human neural system for face perception. Trends Cogn Sci 4(6):223-233. https://doi.org/10.1016/S1364-6613(00)01482-0\u003c/li\u003e\n\u003cli\u003eHaynes JD, Rees G (2006) Decoding mental states from brain activity in humans. Nat Rev Neurosci 7:523-534. https://doi.org/10.1038/nrn1931\u003c/li\u003e\n\u003cli\u003eHougaard A, Jensen BH, Amin FM, Rostrup E, Hoffmann MB, Ashina M (2015) Cerebral Asymmetry of fMRI-BOLD Responses to Visual Stimulation. PLoS ONE 10(5):e0126477. https://doi.org/10.1371/journal.pone.0126477\u003c/li\u003e\n\u003cli\u003eHoward JD, Kahnt T (2021) To be specific: The role of orbitofrontal cortex in signaling reward identity. Behav Neurosci 135(2):210-217. https://doi.org/10.1037/bne0000455\u003c/li\u003e\n\u003cli\u003eHu C, Di X, Eickhoff SB, Zhang M, Peng K, Guo H, Sui J. (2016) Distinct and common aspects of physical and psychological self-representation in the brain: A meta-analysis of self-bias in facial and self-referential judgements. Neurosci Biobehav Rev 61:197-207. https://doi.org/10.1016/j.neubiorev.2015.12.003\u003c/li\u003e\n\u003cli\u003eJabakhanji R, Vigotsky AD, Bielefeld J, Huang L, Baliki MN, Iannetti G, Apkarian AV (2022) Limits of decoding mental states with fMRI. Cortex 149:101-122. https://doi.org/10.1016/j.cortex.2021.12.015\u003c/li\u003e\n\u003cli\u003eJanowska A, Balugas B, Pardillo M et al (2021) The Neurological Asymmetry of Self-Face Recognition. Symmetry 13(7):1135. https://doi.org/10.3390/sym13071135\u003c/li\u003e\n\u003cli\u003eKanwisher N, McDermott J, Chun MM (1997) The fusiform face area: a module in human extrastriate cortex specialized for face perception. J Neurophysiol 17:4302-4311. https://doi.org/10.1523/JNEUROSCI.17-11-04302.1997\u003c/li\u003e\n\u003cli\u003eKnyazev GG, Savostyanov AN, Bocharov AV, Saprigyn AE, Levin EA (2024) Investigating the properties of fMRI-based signature of recognizing one\u0026rsquo;s own face. Biol Psychol 108960. https://doi.org/10.1016/j.biopsycho.2024.108960\u003c/li\u003e\n\u003cli\u003eKoo TK, Li MY (2016) A guideline of selecting and reporting intraclass correlation coefficients for reliability research. J Chiropr Med 15:155\u0026ndash;163. https://doi.org/10.1016/j.jcm.2016.02.012\u003c/li\u003e\n\u003cli\u003eKragel PA, Koban L, Barrett LF, Wager TD (2018) Representation, pattern information, and brain signatures: from neurons to neuroimaging. Neuron 99:257\u0026ndash;273. https://doi.org/10.1016/j.neuron.2018.06.009\u003c/li\u003e\n\u003cli\u003eKriegeskorte N, Mur M, Bandettini PA (2008) Representational similarity analysis-connecting the branches of systems neuroscience. Front Syst Neurosci 2:249. https://doi.org/10.3389/neuro.06.004.2008\u003c/li\u003e\n\u003cli\u003eLevine SC, Banich MT, Koch-Weser MP (1988) Face recognition: a general or specific right hemisphere capacity?. Brain Cogn 8(3):303-325. https://doi.org/10.1016/0278-2626(88)90057-7\u003c/li\u003e\n\u003cli\u003eLyons M, Morikawa K (2020) A Linked Aggregate Code for Processing Faces (Revised Version). arXiv preprint arXiv:2009.08281. https://doi.org/10.48550/arXiv.2009.08281\u003c/li\u003e\n\u003cli\u003eMolnar-Szakacs I, Uddin LQ (2023) Laterality and hemispheric specialization of self-face recognition. Neuropsychologia 108586. https://doi.org/10.1016/j.neuropsychologia.2023.108586\u003c/li\u003e\n\u003cli\u003eMorita T, Tanabe HC, Sasaki AT, Shimada K, Kakigi R, Sadato N (2014) The anterior insular and anterior cingulate cortices in emotional processing for self-face recognition. Soc Cogn Affect Neurosci 9(5):570-579. https://doi.org/10.1093/scan/nst011\u003c/li\u003e\n\u003cli\u003eNili H, Wingfield C, Walther A, Su L, Marslen-Wilson W, Kriegeskorte N (2014) A toolbox for representational similarity analysis. PLoS Comput Biol 10(4):e1003553. https://doi.org/10.1371/journal.pcbi.1003553\u003c/li\u003e\n\u003cli\u003eNorman KA, Polyn SM, Detre GJ, Haxby JV (2006) Beyond mind-reading: multi-voxel pattern analysis of fMRT data. Trends Cogn Sci 10, 424-430. https://doi.org/10.1016/j.tics.2006.07.005\u003c/li\u003e\n\u003cli\u003eNorthoff G, Heinzel A, de Greck M, Bermpohl F, Dobrowolny H, Panksepp J (2006) Self-referential processing in our brain - a meta-analysis of imaging studies on the self. NeuroImage 31(1):440\u0026ndash;457. https://doi.org/10.1016/j.neuroimage.2005.12.002\u003c/li\u003e\n\u003cli\u003eNorthoff G, Qin P, Feinberg TE (2011) Brain imaging of the self \u0026ndash; conceptual, anatomical and methodological issues. Conscious Cogn 20(1):52\u0026ndash;63. https://doi.org/10.1016/j.concog.2010.09.011\u003c/li\u003e\n\u003cli\u003eOosterhof NN, Connolly AC, Haxby JV (2016) CoSMoMVPA: multi-modal multivariate pattern analysis of neuroimaging data in Matlab/GNU Octave. Front Neuroinform 10:27. https://doi.org/10.3389/fninf.2016.00027\u003c/li\u003e\n\u003cli\u003eOta C, Nakano T (2021) Self-face activates the dopamine reward pathway without awareness. Cereb Cortex 31(10):4420-4426. https://doi.org/10.1093/cercor/bhab096\u003c/li\u003e\n\u003cli\u003ePereira F, Mitchell T, Botvinick M (2009) Machine learning classifiers and fMRT: a tutorial overview. NeuroImage 45:199-209. https://doi.org/10.1016/j.neuroimage.2008.11.007\u003c/li\u003e\n\u003cli\u003ePlatek SM, Krill AL, Wilson B (2009) Implicit trustworthiness ratings of self-resembling faces activate brain centers involved in reward. Neuropsychologia 47(1):289-293. https://doi.org/10.1016/j.neuropsychologia.2008.07.018\u003c/li\u003e\n\u003cli\u003ePopal H, Wang Y, Olson IR (2019) A guide to representational similarity analysis for social neuroscience. Soc Cogn Affect Neurosci 14(11):1243-1253. https://doi.org/10.1093/scan/nsz099\u003c/li\u003e\n\u003cli\u003ePourtois G, Schwartz S, Seghier ML, Lazeyras, Vuilleumier P (2005) View-independent coding of face identity in frontal and temporal cortices is modulated by familiarity: an event-related fMRI study. Neuroimage 24(4):1214-1224. https://doi.org/10.1016/j.neuroimage.2004.10.038\u003c/li\u003e\n\u003cli\u003ePrince JS, Charest I, Kurzawski JW, Pyles JA, Tarr MJ, Kay KN (2022) Improving the accuracy of single-trial fMRI response estimates using GLMsingle. Elife 11:e77599. https://doi.org/10.7554/eLife.77599\u003c/li\u003e\n\u003cli\u003eRamon M, Vizioli L, Liu-Shuang J, Rossion B (2015) Neural microgenesis of personally familiar face recognition. PNAS 112(35):E4835-E4844. https://doi.org/10.1073/pnas.1414929112\u003c/li\u003e\n\u003cli\u003eRitchie JB, Kaplan DM, Klein C (2019) Decoding the brain: Neural representation and the limits of multivariate pattern analysis in cognitive neuroscience. Brit J Philos Sci 70:581\u0026ndash;607. https://doi.org/10.1093/bjps/axx023\u003c/li\u003e\n\u003cli\u003eRossion B (2014) Understanding face perception by means of prosopagnosia and neuroimaging. Front Biosci 6:258\u0026ndash;307. https://doi.org/10.2741/e706\u003c/li\u003e\n\u003cli\u003eShrout PE, Bolger N (2002) Mediation in experimental and nonexperimental studies: New procedures and recommendations. Psychol Methods 7:422-445. https://doi.org/10.1037/1082-989X.7.4.422\u003c/li\u003e\n\u003cli\u003eSugiura M, Miyauchi CM, Kotozaki Y et al (2015) Neural mechanism for mirrored self-face recognition. Cereb Cortex 25(9):2806-2814. https://doi.org/10.1093/cercor/bhu077\u003c/li\u003e\n\u003cli\u003eSui J, Humphreys GW (2015) The interaction between self-bias and reward: Evidence for common and distinct processes. Q J Exp Psychol 68(10):1952-1964. https://doi.org/10.1080/17470218.2015.1023207\u003c/li\u003e\n\u003cli\u003eTramacere A (2022) Face yourself: The social neuroscience of mirror gazing. Front Psychol 13:949211. https://doi.org/10.3389/fpsyg.2022.949211\u003c/li\u003e\n\u003cli\u003eUddin LQ, Kaplan JT, Molnar-Szakacs I, Zaidel E, Iacoboni M (2005) Self-face recognition activates a frontoparietal \u0026ldquo;mirror\u0026rdquo; network in the right hemisphere: an event-related fMRI study. NeuroImage 25(3):926-935. https://doi.org/10.1016/j.neuroimage.2004.12.018\u003c/li\u003e\n\u003cli\u003eVigotsky AD, Iannetti GD, Apkarian AV (2024) Mental state decoders: game-changers or wishful thinking?. Trends Cogn Sci 28(10):884-895. https://doi.org/10.1016/j.tics.2024.06.004\u003c/li\u003e\n\u003cli\u003eWager TD, Atlas LY, Leotti LA, Rilling JK (2011) Predicting individual differences in placebo analgesia: contributions of brain activity during anticipation and pain experience. J Neurosci 31:439-452. https://doi.org/10.1523/JNEUROSCI.3420-10.2011\u003c/li\u003e\n\u003cli\u003eWager TD, Atlas LY, Lindquist MA, Roy M, Woo CW, Kross E (2013) An fMRI-based neurologic signature of physical pain. N Engl J Med 368(15):1388-1397. https://doi.org/10.1056/NEJMoa1204471\u003c/li\u003e\n\u003cli\u003eWager TD, Waugh CE, Lindquist M, Noll DC, Fredrickson BL, Taylor SF (2009) Brain mediators of cardiovascular responses to social threat, Part I: Reciprocal dorsal and ventral sub-regions of the medial prefrontal cortex and heart-rate reactivity. NeuroImage 47:821-835. https://doi.org/10.1016/j.neuroimage.2009.05.043\u003c/li\u003e\n\u003cli\u003eWeiskopf DA (2021) Data Mining the Brain to Decode the Mind. In: Calzavarini F, Viola M (eds) Neural Mechanisms. Studies in Brain and Mind, vol 17. Springer, Cham, pp 85-110. https://doi.org/10.1007/978-3-030-54092-0_5\u003c/li\u003e\n\u003cli\u003eWorsley KJ, Marrett S, Neelin P, Vanda AC, Friston KJ, Evans AC (1996) A unified statistical approach or determining significant signals in images of cerebral activation. Hum Brain Mapp 4:58-73. https://doi.org/10.1002/(SICI)1097-0193(1996)4:1\u0026lt;58::AID-HBM4\u0026gt;3.0.CO;2-O\u003c/li\u003e\n\u003cli\u003eYarkoni T, Poldrack RA, Nichols TE, Van Essen DC, Wager TD (2011) Large-scale automated synthesis of human functional neuroimaging data. Nat methods 8(8):665-670. https://doi.org/10.1038/nmeth.1635\u003c/li\u003e\n\u003cli\u003eYue X, Biederman I, Mangini MC, von der Malsburg C, Amir O (2012) Predicting the psychophysical similarity of faces and non-face complex shapes by image-based measures. Vis Res 55:41-46. https://doi.org/10.1016/j.visres.2011.12.012\u003c/li\u003e\n\u003cli\u003eZhan Y, Chen J, Xiao X, Li J, Yang Z, Fan W, Zhong Y (2016) Reward promotes self-face processing: An event-related potential study. Front Psychol 7:735. https://doi.org/10.3389/fpsyg.2016.00735\u003c/li\u003e\n\u003cli\u003eZhou F, Zhao W, Qi Z et al (2021) A distributed fMRI-based signature for the subjective experience of fear. Nat Commun 12(1):6643. https://doi.org/10.1038/s41467-021-26977-3\u003c/li\u003e\n\u003cli\u003eŻochowska A, Nowicka MM, W\u0026oacute;jcik MJ, Nowicka A (2021) Self-face and emotional faces\u0026mdash;are they alike?. Soc Cogn Affect Neurosci 16(6):593-607. https://doi.org/10.1093/scan/nsab020\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1. Prediction-outcome correlations for ROI-based signatures in the \u0026lsquo;self\u0026rsquo; and \u0026lsquo;friend\u0026rsquo; conditions.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eROI\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lsquo;self\u0026rsquo;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lsquo;friend\u0026rsquo;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003eEarly visual\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e0.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003eInferior occipital\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e0.34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e0.32\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003eFusiform face area\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e0.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003eFace recognition\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e0.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e0.16\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003eSelf-referential\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e0.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e0.20\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003eMedial orbitofrontal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e0.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33.3333%;\"\u003e\n \u003cp\u003e0.09\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eAll correlations are significant at p \u0026lt; 0.05, FDR-corrected.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 2. RSA results. Spearman correlations between ROI RDMs and behavioral RDMs averaged across participants. Significance of correlations was tested using a one-sample permutation t-test across participants.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 36.7347%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eROI\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30.6122%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lsquo;self\u0026rsquo;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32.6531%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lsquo;friend\u0026rsquo;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 36.7347%;\"\u003e\n \u003cp\u003eEarly visual\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30.6122%;\"\u003e\n \u003cp\u003e0.64*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32.6531%;\"\u003e\n \u003cp\u003e0.62*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 36.7347%;\"\u003e\n \u003cp\u003eInferior occipital\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30.6122%;\"\u003e\n \u003cp\u003e0.22*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32.6531%;\"\u003e\n \u003cp\u003e0.23*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 36.7347%;\"\u003e\n \u003cp\u003eFusiform face area\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30.6122%;\"\u003e\n \u003cp\u003e0.21*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32.6531%;\"\u003e\n \u003cp\u003e0.20*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 36.7347%;\"\u003e\n \u003cp\u003eFace recognition\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30.6122%;\"\u003e\n \u003cp\u003e0.14*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32.6531%;\"\u003e\n \u003cp\u003e0.12*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 36.7347%;\"\u003e\n \u003cp\u003eSelf-referential\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30.6122%;\"\u003e\n \u003cp\u003e0.25*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32.6531%;\"\u003e\n \u003cp\u003e0.13*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 36.7347%;\"\u003e\n \u003cp\u003eMedial orbitofrontal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30.6122%;\"\u003e\n \u003cp\u003e0.08*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32.6531%;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e* correlation is significant at p \u0026lt; 0.05, FDR-corrected.\u0026nbsp;\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"biomarker, multivariate pattern analysis, neural signature, physical self, face similarity","lastPublishedDoi":"10.21203/rs.3.rs-7570992/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7570992/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eFacial similarity to one\u0026rsquo;s own face indirectly affects a person\u0026rsquo;s social behavior. In this study, we aimed to develop, using multivariate pattern analysis, a brain signature for assessing facial similarity to one\u0026rsquo;s own face (SFSS). We hypothesized that it should have at least two aspects: one related to the assessment of similarity of faces, and the other specifically to the assessment of similarity to one\u0026rsquo;s own face. To distinguish between these aspects, we included two tasks that used the same set of morphed images, but in one case the task was to assess similarity to one\u0026rsquo;s own face and in the other to another person\u0026rsquo;s face. The SFSS showed excellent correlation with behavioral data and a large effect size in predicting the instantaneous similarity score for each participant. Mediation analysis showed that it acts as a mediator between the degree of similarity objectively present in an image and the subjective similarity assessment. Principal component analysis allowed to separate brain regions associated with the assessment of face similarity in general and with judging the similarity to one\u0026rsquo;s own face. It appeared that the early visual cortex was particularly involved in the former, while the fusiform gyrus and orbitofrontal cortex were involved in distinguishing targets. The results show the fruitfulness of considering brain signatures as representations of complex psychological processes, which can be analyzed by decomposing them into components to represent different aspects of relevant psychological states.\u003c/p\u003e","manuscriptTitle":"A Neural Signature of Similarity Assessment between a Face and One’s Own Face","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-21 23:07:49","doi":"10.21203/rs.3.rs-7570992/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"5b2aa2c0-9c35-40d0-b712-b517b836d30e","owner":[],"postedDate":"October 21st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-02-12T03:10:26+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-21 23:07:49","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7570992","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7570992","identity":"rs-7570992","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}