Community Research and Development Information Service - CORDIS


PINS Report Summary

Project ID: 629004
Funded under: FP7-PEOPLE
Country: United Kingdom

Periodic Report Summary 1 - PINS (The neural representations of person identity information: fMRI and neuropsychological investigations)

Project description:
This project aims to investigate the cognitive and neural bases of person identity nodes (PINs). According to an influential model of face perception (Bruce & Young, 1986), faces are processed in stages. In this model, visual information derived from faces is first used to encode facial structure, after which the structural information is passed on to face recognition units (FRUs), where structural codes of previously seen faces are stored. Faces are successfully recognised when visual information matches a structural code stored in the FRUs. Similar modality-dependent recognition units can be conceptualised for voices, names, or other cues typically used for person recognition (e.g., Belin et al., 2011). FRUs then send the face information to PINs, which contain or provide access to semantic, modality-independent information associated with the person. However, despite the model’s intuitive appeal, there is limited evidence supporting the existence of these abstract, modality-independent modes. Our project was proposed to fill this critical gap by investigating human brain representations of abstract (i.e. extracted away from sensory information) identity information. Furthermore, we aim to examine the modularity of the person identity recognition system (whether face recognition units are independent of the voice recognition units) and whether PINS are affected when domain-specific recognition systems are compromised due to atypical neurodevelopment.

Work performed and main results:
We have performed three fMRI experiments aimed at identifying and validating brain areas that can reliably distinguish between person identities across dramatically different sensory inputs. For this purpose, we decided to focus on visual stimuli instead of combining stimuli from various sensory modalities. Previous studies showed that neural activation in certain brain areas can predict with above chance accuracy the identity we see in natural (i.e. photographic) face views, but the effects were small despite the very similar visual stimuli were used (identical faces, Kriegeskorte et al., 2007; or faces varying only in viewpoint or expression, Anzellotti et al., 2013; Nestor et al., 2011). Therefore, the next step was to establish if there are brain areas with a reliable response to person identity when stimuli vary dramatically within the same modality before taking the investigation across modalities.
Our aim was to investigate whether the results found with realistic photographs will hold for other visual representations of faces, such as caricatures. This is important because realistic photos of different individuals (e.g. Brad Pitt and George Clooney) typically share more physical similarities than a photo and a caricature of the same individual. Brain areas responding more similarly to person identity (e.g. photo and caricature of Brad Pitt) rather than image format are therefore likely to encode a more abstract representation of identity, not dependant on a realistic depiction of facial features.
To create our stimuli, we contacted a local artist to draw caricatures of three famous actors (Brad Pitt, George Clooney, and Tom Cruise) based on photos presenting their faces from three different viewpoints: frontal, side, and full profile. In addition, the hired artist also created line drawings of the three actors – black and white minimalistic sketches true to the real facial structure (i.e. without exaggerating the features). The photos, line drawings and caricatures were the stimuli used in our experiments (Figure 1).

Figure 1. Frontal, side and profile face shots of three actors and the corresponding line drawings and caricatures created by an artist.

For the first experiment, we had 13 participants undergoing a short training session until they could reliably recognise the three actors from any image, and then undergoing fMRI scanning. Their task in the scanner was to press a button whenever they recognised Tom Cruise. The trials presenting Tom Cruise were in fact filler trials, because for data analysis we were interested only in the trials with images of Brad Pitt and George Clooney. Our aim was to identify the brain areas where neural activation patterns could reliably predict whether participants were looking at Brad Pitt or George Clooney regardless of image type. For data analysis, we used multivariate pattern analysis (MVPA; Haxby et al., 2001), a method that focuses on patterns of activation across voxels. Linear support vector machine (SVM) classifiers were trained on neural activation patterns in regions of interest and performance was measured by how well the classifier predicted category membership of new data. In fMRI analysis, successful classifiers indicate that a particular region is sensitive to the experimental conditions.
Brain regions of interest were identified with a standard face localiser task. These regions of interest were: the fusiform face area (FFA), the occipital face area (OFA), the superior temporal sulcus (STS) and the face-selective anterior temporal lobe (fATL). We used the early visual cortex (EVC) as a control region of interest. Confirming that our design was appropriate to detect significant effects, we found that image type (i.e. photos, line drawings or caricatures) can be reliably distinguished in EVC, an area known to respond to the physical attributes of visual information. OFA, FFA, and STS were also sensitive to the image format. Most importantly for our study, when we looked at brain areas encoding identity regardless of image type (e.g. Brad Pitt from photos, line drawings, and caricatures), we found classification accuracy significantly above chance in STS (when we considered regions of interest with 6mm and 9mm radii) and FFA (when we considered regions of interest with 6mm radius) (Figure 2).

Figure 2. Accuracy of person identity classification in our regions of interest. Significant results are marked with a red asterisk. rV1=right early visual cortex, rOFA=right occipital face area, rSTS=right superior temporal sulcus, rFFA=right fusiform face area, rATL=right anterior temporal lobe.

The low power and limited replications of fMRI studies (especially those reporting subtle brain effects) is a serious issue permeating neuroimaging research. For example, previous findings related to representations of face identity were based on results from only 8 (Nestor et al., 2011), 9 (Anzellotti et al., 2013) or 11 participants (Kriegeskorte et al., 2007). Our findings were based on results from 11 participants (two participants were discarded because of poor data quality), and the reported effects were small (in line with previous studies). Therefore, our next goal was to confirm the robustness of our initial results, and this took precedence over running another novel but under-powered fMRI study.
Our next two experiments were aimed at replicating the results of our initial study. To increase the statistical power to detect meaningful results, we increased the number of presentations per image type and the number of participants. The follow up experiments included only photos and caricatures of the same three actors (Brad Pitt, George Clooney, and Tom Cruise); line drawings were dropped so that we could have more repetitions for photos and caricatures. We had 15 participants in the first follow-up experiment and 15 participants in the second follow-up experiment. The results provided a strong confirmation of our initial findings that the right posterior STS is involved in abstract representations of person identity; the neural pattern activities to images of Brad Pitt and George Clooney differed in predictable ways regardless of whether photos or caricatures of the actors were presented. This finding is consistent with recent data (Anzellotti et al., 2015) finding multimodal representations of identity (using faces and voices) in posterior STS. Furthermore, we have obtained some evidence (though not as strong as for the right posterior STS) that the right FFA and ATL may also be involved in encoding abstract representations of person identities.

Expected final results and impact:
Our three fMRI experiments represent a comprehensive and robust investigation of abstract representations of person identity from visual input. Although photographs of different people are more similar in terms of low-level visual information than a photograph and a caricature of the same person, we found some brain regions (most consistently the posterior STS, but also possibly the FFA and the ATL) that care more about the identity of the person in the image than the type of the image. This is a first indication about the existence of abstract person identity nodes.
In the remaining year of this project, we will examine the modularity of the person identity recognition system. Specifically, we plan to test a relatively large number of individuals with developmental prosopagnosia (the inability to recognise faces normally in the absence of any known brain damage) on voice recognition. Normal voice recognition in these individuals will suggest that the face recognition units are independent of the voice recognition units. In contrast, abnormal voice recognition will suggest one of two possibilities: i) the domain-specific recognition systems depend on each other, or ii) atypical neurodevelopment of domain-specific recognition systems (i.e. FRUs) may affect the normal development of the abstract identity nodes (i.e. PINS) which affect recognition of person identity from other modalities (in our case voices). We have already started preliminary testing of our large database of developmental prosopagnosics (from to select a homegeneous group of participants (e.g. with “pure” prosopagnosia) to be included in our study.
Our work will contribute to contemporary theories of person identity recognition. Basic knowledge about the neural mechanisms of person recognition may lay the groundwork for translational research into rehabilitative strategies for individuals with atypical person recognition abilities.

Reported by

United Kingdom


Life Sciences
Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top