Beyond the FFA: Brain-behavior correspondences in face recognition abilities
Introduction
Despite the thousands of papers dedicated to understanding the neural basis of face perception in both humans and non-human primates, very little is known about how neural activation relates to the multifaceted components of face processing behavior. The little that is known about brain-behavior correspondences in the face-processing system has largely focused on relating behavior to activation in the pre-eminent fusiform face area (FFA), what is now known as FFA1 (Grill-Spector and Weiner 2014). However, much of this work has produced inconsistent findings and is prone to alternative interpretations. Here, we systematically characterized the full range of individual differences in face recognition behavior and in brain-behavior correspondences within the set of distributed nodes of the face-processing network in healthy typically developing (TD) adults.
Studies employing an individual differences approach, that is examining associations and dissociations in behavior across individuals, can answer fundamental questions about the way face processing operates (Yovel et al. 2014). This approach has been used to investigate brain-behavior relations in the face-processing system, with a strong focus on the right FFA (e.g., Weibert and Andrews, 2015). For example, Grill-Spector and colleagues were the first to report a parametric increase in activation in the right FFA as a function of recognition behavior in 5 individuals (Grill-Spector et al. 2004). Three papers reported a positive correlation between the behavioral face inversion effect (i.e., upright>inverted accuracy) and neural face inversion effect (FIE: upright> inverted; Aylward et al. 2005; Yovel and Kanwisher 2005; Passarotti et al. 2007) in the right fusiform gyrus (FG). However, age was not controlled for in two of these experiments (Aylward et al., 2005, Passarotti et al., 2007) and participants ranged in age from 8 to 30 years. As a result, “individual differences” in the relation between the behavioral and neural FIEs may have been largely age-related. More recently, Furl et al. (2011) observed a positive association between face recognition accuracy and the size and magnitude of activation in the FFA in a combined sample of congenital prosopagnosics and typically developing adults. However, they did not report whether such relations held in either of the populations separately. Huang and colleagues reported a positive association between the magnitude of activation in the right FFA and face recognition behavior (Huang et al. 2014). Although the participants were Chinese young adults, the researchers defined face-selectivity based on responses to Caucasian child faces and did not quantify differential experience with Caucasian people among the participants. As a result, individual differences in experience with Caucasian faces may have contributed to differences in the magnitude of the responses in the FFA. Therefore, these findings are all prone to alternative interpretations regarding the mechanism of the individual differences; they could be modulated by age, patient status, or experience factors related to behavior.
In contrast, other researchers have reported difficulty observing a relation between individual differences in face recognition behavior and FFA selectively for faces (see McGugin and Gauthier, 2015; McGugin et al. 2016), suggesting that adults may be at ceiling with respect to their behavioral visuoperceptual expertise for faces. Finally, congenital prosopagnosics who are face blind exhibit perfectly normal magnitude activation in the FFA (Avidan et al. 2005). In sum, the work relating face recognition behavior to neural activation is variable, narrowly focused on the right FFA, and prone to alternative interpretations. The current work addresses these limitations using a multipronged approach.
First, we propose that there is a wide range of individual differences in face recognition behavior among typically developing individuals (Wilmer et al., 2010). Importantly, observation of these individual differences requires careful screening for a personal and/or family history of psychiatric and/or neurological illness because face processing is effected by every social-emotional disorder (Scherf et al., 2008, Avidan et al., 2005, Davis et al., 2011, Calkins et al., 2005). In previous studies, failure to exclude participants who have genetic vulnerabilities for these disorders may have masked the observation of typical individual differences. Here, we assessed a large sample of healthy, typically developing adults who had no history of psychiatric or neurologic illness in themselves or their first-degree relatives in a battery of face and object recognition tasks. This allowed us to observe a full range of individual differences in recognition behavior among typically developing adults.
Second, we evaluated brain-behavior relations throughout the face-processing network, including both core (FFA, OFA, pSTS) and extended regions like the amygdala, PCC, vmPFC, and ATL (Gobbini and Haxby, 2007). We also defined separate face patches within the fusiform gyri bilaterally (see Weiner et al., 2014) for each participant to evaluate the functional contributions of FFA1, FFA2, and the posterior face patch (pFG/IOG) to face recognition abilities. In so doing, we built upon an emerging perspective that the seemingly independent components of face perception that have been previously ascribed to individual regions, like recognition in the FFA (Kanwisher et al. 1997), emerge from interactions among many distributed regions (Cohen Kadosh et al., 2011, Ishai, 2008, Nestor et al., 2011, Vuilleumier and Pourtois, 2007).
Third, we identified a set of behavioral and neural measures to capture a multifaceted perspective on assessing brain-behavior relations. Specifically, we employed convergent measures of face recognition behavior that engage familiar and unfamiliar face recognition. We also engaged object recognition behavior to determine the specificity of the brain-behavior relations. With respect to the neural measures, we included the standard measure of activation at the region level (i.e., magnitude of activation). Increases in magnitude are generally interpreted to reflect an increase in neural resources and are likely related to increases in the firing rate of neurons and/or the number of neurons firing. Support for this in the face-processing system comes from recent work in which simultaneous electrocorticography (Ecog) and fMRI recordings in the same participants reveal an increase in the firing rate of populations of neurons in the fusiform gyrus at particular frequencies when individuals look at faces, which is correlated with face-selective responses in the fMRI BOLD signal (Jacques et al. 2016).
In addition to magnitude, we also measured the size of activation for each individually defined ROI (i.e., number of significant contiguously active voxels). When functional data are not smoothed, the size of the region can provide information about extent of the local distributed representation (Scherf et al., 2007, Golarai et al., 2010, Weiner and Grill-Spector, 2011). Previous findings from developmental neuroimaging studies on the emerging topography in the ventral visual pathway have reported age-related increases in the size of the functionally defined FFA and OFA from childhood to adulthood (e.g., Scherf et al. 2007; Golarai et al. 2010). Furthermore, these increases in the size of the regions, but not the magnitude, correspond with an improvement in face recognition behavior (Golarai et al. 2010). These findings suggest that the size of activation may be a particularly sensitive measure for tracking individual differences in brain-behavior relations within the face-processing system. It may reflect the integration of local circuits that carry information about distributed representations, which manifests as larger functional regions. Finally, we also included a neural measure that represented network level activation. This measure reflects the notion that complex behavior might be captured by distributed activation across the nodes of the network better than activation within any one specific node.
The central hypothesis is that individuals with better face recognition abilities would dedicate more neural resources to the task of perceiving faces. This would be evident in particularly critical neural regions in terms of the magnitude and size measures. For example, we predicted that we would observe a positive relation between face recognition performance and neural magnitude in multiple regions of the face processing network, including but not limited to the right FFA1. Given previous developmental findings reporting an increase in the size of activation with increasing age (Scherf et al., 2007, Scherf et al., 2014; Golarai et al. 2007) and face recognition ability (Golarai et al. 2010), we predicted that there would be a strong relation between variations in face recognition behavior and the size of activation within face-selective functional regions. In other words, individuals with stronger face recognition skills may exhibit larger face-related functional regions, but not object- or place-related regions. This relation could reflect local circuitry that looks more homogenous at the voxel level because it is integrating information across sparsely distributed representations, which facilitates recognition. Finally, we predicted that superior recognizers would also engage larger proportions of the face-processing network, which would reflect that they are accessing more distributed representations that spread across regions, which support the invariant and semantic components of faces representations.
Section snippets
Participants
Typically developing young adults (N=266, range=18–25 years, 162 females) participated in the behavioral portion of the experiment. Participants were healthy and had no history of neurological or psychiatric disorders in themselves or their first-degree relatives. They were also screened for behavioral symptoms indicative of undiagnosed psychopathology.
We specifically sought to enroll 40 participants (20 male, 20 female) for the neuroimaging study given the sample size recommendations for
Individual differences in behavior
Fig. 2a shows the distribution of accuracy scores for the CFMT+ from the large sample of 266 participants. Five participants scored in the super recognizer range (Russell et al. 2009) and 16 performed in the prosopagnosic range (Duchaine and Nakayama 2006); only 1 of these 16 individuals reported having a personal history of difficulty recognizing faces. Fig. 2b shows the distribution of accuracy scores on the FBF. Performance between the CFMT+ and FBF was strongly related within individuals, F
Discussion
We systematically characterized a broad range of individual differences in face recognition behavior and in brain-behavior correspondences within the nodes of the face-processing network in healthy typically developing (TD) adults. We took a multifaceted approach to quantifying brain-behavior relations within this complex neurocognitive system. To do so, we included the traditional fMRI measure that represents activation (i.e., magnitude), but we also implemented additional measures including
Conclusions
To summarize, with sufficient power and rigorous identification of activation at the individual subject level, we provide the first evidence that activation within multiple regions in the face-processing network, beyond right FFA1, are systematically related to behavior. This includes multiple core and extended regions. Second, we report that variations in face recognition behavior are selectively related to variations in the size of activation more often than variations in the magnitude of
Acknowledgments
K. Suzanne Scherf, Departments of Psychology and Neuroscience, and Daniel Elbich, Department of Psychology, Pennsylvania State University.
The research reported in this paper was supported by The Department of Psychology and the Social Science Research Institute at Penn State University.
We would like to thank Debra Weston from the Social, Life, and Engineering Sciences Center (SLEIC) for her help in acquiring the imaging data, and Sara Barth, Natalie Garcia, Giorgia Picci, Ashley Unger, Mikayla
References (71)
Neural systems for recognizing emotion
Curr. Opin. Neurobiol.
(2002)- et al.
Function and structure of human left fusiform cortex are closely associated with perceptual learning of faces
Curr. Biol.
(2014) - et al.
The Cambridge face memory test: results for neurologically intact individuals and an investigation of its validity using inverted face stimuli and prosopagnosic participants
Neuropsychologia
(2006) - et al.
Probabilistic atlases for face and biological motion perception: an analysis of their reliability and overlap
Neuroimage
(2013) - et al.
The nature of face representations in subcortical regions
Neuropsychologia
(2014) - et al.
Thresholding of statistical maps in functional neuroimaging using the false discovery rate
NeuroImage
(2002) - et al.
Neural systems for recognition of familiar faces
Neuropsychologia
(2007) - et al.
Intelligence and changes in regional cerebral glucose metabolic rate following learning
Intelligence
(1992) - et al.
Neural adaptation across viewpoint and exemplar in fusiform cortex
Brain Cogn.
(2012) - et al.
The distributed human neural system for face perception
Trends Cogn. Sci.
(2000)