Comparing the sensitivity of face matching assessments to detect face perception impairments
Introduction
Faces convey signals essential for social interactions and are one of the most reliable ways to determine a person's identity. Recognizing a face is a highly specialized, multistage process (Bruce and Young, 1986) involving a network of brain regions (Grill-Spector et al., 2017; Haxby et al., 2000; Haxby and Gobbini, 2011). Though recognition of familiar faces is typically rapid and effortless (Jenkins et al., 2018), this ability can be severely impaired in many neurological, psychiatric, and developmental disorders including prosopagnosia (Albonico and Barton, 2019; Mayer and Rossion, 2007), autism spectrum disorders (Dwyer et al., 2019; Weigelt et al., 2012), schizophrenia (Watson, 2013), Alzheimer's disease (Lavallée et al., 2016), person recognition disorders (Gainotti, 2007), age-related cognitive decline (Boutet et al., 2015), as well as others (Barton et al., 2004; Dimitriou et al., 2015). The causes of face recognition deficits in these disorders vary, and it is crucial to identify measures that can discern the stages of face identification impairments to help with both diagnosis (Benton and Van Allen, 1968; Benton et al., 1994; Duchaine and Nakayama, 2006; Duchaine et al., 2007) and treatment (Bate and Bennetts, 2014; DeGutis et al., 2014a, b).
When assessing face recognition abilities, it is important to consider there are two stages of processing involved: face perception and face memory (e.g., De Renzi et al., 1991; Liu et al., 2002; Weigelt et al., 2014). Face perception refers to building up/encoding a structural representation of a face (Bruce and Young, 1986). This representation allows one to judge whether simultaneously presented faces belong to the same or different individuals. On the other hand, face memory involves the ability to store and retrieve individuated faces from long-term memory. Face memory relies on face perception but also involves processes such as associating semantic and contextual information with a face, storing and retrieving a face and related semantic/contextual information, and, in the case of familiar faces, building up a robust face representation over repeated presentations. One influential developmental study suggests that face perception develops earlier (∼age 5) while face memory demonstrates later face-specific development (∼age 10, Weigelt et al., 2014). Additionally, patient studies have found dissociations between acquired prosopagnosics with impaired face perception and intact face memory (apperceptive prosopagnosia) and vice versa (associative prosopagnosia, De Renzi et al., 1991; though see Busigny et al., 2014).
Face perception impairments are typically assessed using face matching tasks where the to-be-matched faces are presented simultaneously and, importantly, vary in either viewpoint, lighting, or emotion (Benton and Van Allen, 1968; Duchaine et al., 2007; White et al., 2017). Such variations in face images prevent direct image-based matching, are thought to make judgments rely more on specialized face-specific perceptual mechanisms (e.g., holistic processing, McKone, 2008; Tanaka and Farah, 1993), and better capture impairments in patients with face perception deficits. Face memory is assessed using tasks that involve learning and short-term retention of identities of novel faces, such as the Cambridge Face Memory Test (CFMT: Duchaine and Nakayama, 2006) or long-term recognition of familiar/famous faces (Bobak et al., 2017; Duchaine and Nakayama, 2005). Although many researchers have regarded the CFMT as the gold standard test for assessing deficits in face memory, there is currently no widely agreed-upon measure to reliably characterize face perception impairments. Notably, because perception precedes memory, all visual memory tests including the CFMT depend on both perceptual and memory processes, thus making it difficult to dissociate the independent contributions of the two. For example, studies have reported that factors that impair face matching performance (e.g., lighting, viewpoint changes) similarly impair short-term and long-term memory for faces (Braje et al., 1998; Braje, 2003). The goal of the current study was to identify tests that can best assess face perception in impaired samples.
In the past two decades, numerous tests have been developed to assess face perception (e.g., Duchaine et al., 2007; Burton et al., 2010; Fysh and Bindemann, 2018). The Benton Face Recognition Test1 (BFRT, Benton and Van Allen, 1968) was one of the first assessments to provide a standardized measure of face perception proficiency to assess deficits. The BFRT requires matching the identity of a front-view target face to three of six faces simultaneously presented, which may vary in lighting or viewpoint, (see Fig. 1). This test has been widely used to assess perceptual deficits in clinical disorders like cortical blindness, lobectomy, visual agnosia, and autism (Benton et al., 1994; Busigny et al., 2009; Demirci and Erdogan, 2016; Duchaine, 2000; Yerys et al., 2018). However, the sensitivity of the original, unspeeded version of the BFRT has been challenged in studies of developmental prosopagnosics (DPs), individuals with severe lifelong impairments in face recognition and group-level face perception deficits (Duchaine and Nakayama, 2004; Duchaine and Weidenfeld, 2003; Nunn et al., 2001). In particular, Duchaine and Nakayama (2004) found that 9 of 11 DPs performed normally on the BFRT (>41 out of 54, z-score > −1). Further, a recent study reported that in 23 DPs, 18 scored within the normal range on the original BFRT (Albonico et al., 2017).
One potentially critical factor with the original version of the BFRT is that it had no time restrictions and did not emphasize response speed. Studies have shown that when given unlimited time to perform face matching, feature-by-feature comparison becomes an available strategy to achieve high accuracy rates (e.g., Towler et al., 2017). Along these lines, studies have reported that prosopagnosics take significantly longer to complete the original BFRT than controls (Albonico et al., 2017; Duchaine, 2000; Duchaine and Nakayama, 2004; Nunn et al., 2001), as they may use a feature-by-feature strategy to achieve accuracy within the normal range. This suggests that speeded tasks may be better at identifying impairments in face perception. In an effort to address these shortcomings of the original BFRT, Rossion and Michel (2018) developed a computerized version of the BFRT (BFRT-c) that collects response times and instructs participants to respond, “as quickly and accurately as possible”. The normative accuracy of this version (see Dzhelyova et al., 2020) is notably 0.75 SDs lower than versions without speeded instructions (e.g., Albonico et al., 2017). This suggests that the BFRT-c's speeded instructions may change participants' strategy, possibly favoring face-specific processes (e.g., holistic processing) rather than more laborious, non-face specific feature matching processes. This may make the BFRT-c a substantially more sensitive assessment of face perception impairment than the original version or other currently available assessments.
To date, there have been no studies comparing the validity and sensitivity of this version of the BFRT-c or other available face perception tests in clinical samples. DPs are an ideal population to validate face perception tests, as larger samples of DPs consistently show reduced group-level face matching performance compared to controls (e.g., Biotti et al., 2019; Dalrymple et al., 2014; Zhao et al., 2016; though individual DP cases may demonstrate normal face perception) and are prevalent enough to obtain adequate group sample sizes (2.5% of population; Kennerknecht et al., 2006). In addition to differentiating between DPs and controls, a good face perception assessment should also strongly correlate with measures of face memory, since face perception ability is a significant contributor to face memory performance (e.g., Stumps et al., 2020). In the current study, we compared the BFRT-c and the widely used Cambridge Face Perception Test (CFPT, Duchaine et al., 2007; Bobak et al., 2017; Bowles et al., 2009; Gonzalez-Perez et al., 2019; Palermo et al., 2017; Rezlescu et al., 2017), in their ability to a) detect face perception impairments and b) predict face memory performance. In the CFPT, participants are required to arrange six front-view morphed faces from most-to-least similar to a target face shown from a ¾-viewpoint. Two important differences between CFPT and BFRT-c are that a) speed is emphasized in the BFRT-c whereas in the CFPT there is less emphasis on speed, with 60 s provided to complete each trial, and b) the BFRT-c includes lighting and viewpoint changes between faces whereas the CFPT only includes viewpoint change trials. Studies have suggested that prosopagnosics may be particularly worse at matching faces across lighting change trials, even more so than viewpoint change trials (Duchaine and Nakayama, 2004; Rossion and Michel, 2018). Given that BFRT-c is speeded and includes lighting change trials, we hypothesized that the BFRT-c will predict unique variance from the CFPT in DP vs. control group membership as well as objective and subjective face recognition ability.
In exploratory analyses, we also sought to compare the BFRT-c and CFPT to two additional face matching assessments shown to be sensitive to face perception impairments. First, we included the USC Face Perception Test (USCFPT, Margalit et al., 2016; Biederman et al., 2017) that uses computer-generated faces and difficulty levels scaled according to the Gabor-jet model and has shown decreased performance in DPs (Biederman et al., 2017; Margalit et al., 2016; Yue et al., 2012). We also included a novel same/different face matching task (SDFMT) (motivated by White et al., 2017), which demonstrated significant differences between DPs and controls. Finally, in the tests with lighting and viewpoint change trials (BFRT-c and SDFMT), we also sought to directly compare these trial types in their ability to differentiate DPs and controls and predict objective and subjective face recognition.
Section snippets
Participants
The study included 60 adults between the ages of 18 and 70 years old. Controls and DPs were recruited from different sources. DPs were recruited from: a) Our database of Boston DPs who previously participated in laboratory studies, b) referrals from Dr. Matthew Peterson at MIT, who recently completed a Boston-area DP study (Peterson et al., 2019), c) referrals from Dr. Brad Duchaine's website, www.faceblind.org, and d) responses to our advertisement posted on public transportation
Demographics and diagnostic test performance
Participants included 30 DPs (24 female) and 30 controls (18 female) with a mean age of 38.5 years (SD = 13.69) and 38.8 years (SD = 10.18), respectively. There was no significant difference between the two groups in age (p = .92), but there was a trend toward a higher proportion of females in the DP group (p = .09). As expected, the DP group performed significantly worse (all p < .001) than controls on objective and subjective diagnostic measures of face recognition (CFMT, FFMT, and PI20;
Discussion
The current study helps to address the existing gap in the literature regarding which tests are best for identifying face perception impairments. When diagnosing DP as well as predicting objective and subjective face recognition, the BFRT-c and CFPT consistently showed robust predictive abilities and clearly outperformed the other measures. When we directly compared the BFRT-c and CFPT, the BFRT-c outperformed the CFPT, solely predicting unique variance in DP diagnosis and self-reported face
Conclusion
Research has shown that face perception and face memory are interconnected but distinct processes. Yet, in contrast to face memory assessments that are better-established, studies have yet to examine the best assessments to reliably measure face perception impairments. Encoding faces accurately is an initial step in developing face familiarity and later stages of face memory, person recognition, and integration with subsequent cognitive processes. The current study examines multiple face
Credit author statement
Maruti Mishra: Conceptualization, Data Curation, Formal analysis, Visualization, Writing - Original Draft, Writing - Review & Editing. Regan Fry: Investigation, Data curation, Writing- Review & Editing. Elyana Saad: Methodology, Software. Joseph Arizpe: Methodology, Software, Writing- Review & Editing. Yuri-Grace Ohashi: Investigation. Joseph M DeGutis: Conceptualization, Methodology, Resources, Writing - Original Draft, Writing - Review & Editing, Supervision, Project administration, Funding
Declaration of competing interest
We have no known conflicts of interest to disclose.
Acknowledgement
We thank Alice E. Lee and volunteers at HDSL for data collection of DP and control participants, respectively. We want to thank developmental prosopagnosics and control participants for completing our challenging battery of tasks.
References (95)
- et al.
Impairment in facial recognition in patients with cerebral disease
Cortex
(1968) - et al.
Prosopagnosia and facial discrimination
J. Neurol. Sci.
(1972) - et al.
Is developmental prosopagnosia best characterized as an apperceptive or mnemonic condition?
Neuropsychologia
(2019) - et al.
Face-specific impairment in holistic perception following focal lesion of the right anterior temporal lobe
Neuropsychologia
(2014) - et al.
Dissociation between face perception and face memory in adults, but not children, with developmental prosopagnosia
Developmental Cognitive Neuroscience
(2014) - et al.
Apperceptive and associative forms of prosopagnosia
Cortex
(1991) - et al.
Evidence for perceptual deficits in associative visual (prosop)agnosia: a single-case study
Neuropsychologia
(2004) - et al.
An evaluation of two commonly used tests of unfamiliar face recognition
Neuropsychologia
(2003) - 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.
Investigating the perception of face identity in adults on the autism spectrum using behavioural and electrophysiological measures
Vis. Res.
(2019)