A hallmark of the human visual system is our ability to make rapid visual categorizations in fractions of a second, whether we are interpreting the meaning of a picture (Potter, Wyble, Hagmann, & McCourt, 2013), classifying a scene (Schyns & Oliva, 1994) or recognizing a familiar face (Grill-Spector & Kanwisher, 2005).
In a medical evaluation, diagnosis of chest radiographs and mammograms requires the detection and localization of the radiological abnormality (Kundel, Nodine, Conant, & Weinstein, 2007). After an initial glimpse, expert radiologists report that they have an intuition that a mammogram is likely to be normal or abnormal before any pathology is localized. In searching for signs of lung cancer, Kundel and Nodine (1975) found that radiologists could achieve a d’ of about 1.0 after a just 200 ms glimpse of a chest X-ray. This level of performance was nowhere near the level of d’ of 2.5 obtained during free-viewing conditions of these stimuli but, nevertheless, comfortably above chance performance. In mammography, Evans, Georgian-Smith, Tambouret, Birdwell, and Wolfe (2013) found a similar level of performance after a 250 ms exposure to mammograms. The rapid global analysis of the radiological image has been referred to as holistic processing and is the precursor to the subsequent stage involved in the localization of abnormalities (Carrigan et al., 2018).
In holistic processing, recognition relies on the integration of individual stimulus parts into an emergent whole representation that is qualitatively more than the summed representation of its individual parts. Face recognition is the prime example of holistic processing where recognition is based on the synthesis of facial features that yields a unique face that is more than the summed recognition of each individual facial feature. Three tasks have been applied as the gold standards for testing holistic face processes: the face composite task, the parts/wholes task, and the inversion task. In the “face composite task”, it has been demonstrated that participants find it difficult to selectively attend to one half of a face (e.g., top half) while ignoring information from the other half (e.g., bottom half) (Young, Hellawell, & Hay, 1987). In the face composite task, the whole face representation makes it difficult for participants to selectively attend to one region of the face, isolated from the whole face. In the “parts/wholes task”, participants exhibit better recognition when a face part (mouth) is displayed in the whole face than when displayed in isolation (Tanaka & Farah, 1993; reviewed in Tanaka & Simonyi, 2016). The parts/wholes task demonstrates that facial features are not represented in memory as individual parts, but are integrated into a whole face representation.
Perhaps the most widely used test of holistic face processes is the face inversion task (Yin, 1969). Although all objects are more difficult to recognize when inverted compared to upright, inversion disproportionately impairs the recognition of faces relative to other object classes (McKone & Yovel, 2009; Rossion, 2008; Yin, 1969). Turning a face upside down disrupts the normal holistic face processing and forces the participant to use a less optimal strategy based on analysis of specific features (wide-set eyes, square jaw, etc). Inversion has been shown to abolish the holistic interference observed in both the face composite task (Rossion & Boremanse, 2008; Young et al., 1987) and the whole face recognition advantage in the parts/whole task (Tanaka & Farah, 1993; Tanaka & Sengco, 1997).
Real world perceptual experts, such as birdwatchers or dog judges, are similar to face “experts” in that they recognize objects in their domain of expertise quickly, accurately, and at a specific level of categorization (Tanaka & Taylor, 1991). To facilitate their speeded precognition, it has been hypothesized that expert recognition demands the same kind of holistic processing that is employed in face processing. Therefore, it follows that expert object recognition should be susceptible to similar manipulations used in face recognition, such as inversion. In a seminal study, Diamond and Carey (1968) tested this prediction by asking dog judges and control participants to recognize upright and inverted photographs of dogs. They found that while the novices exhibited an inversion effect only for faces, dog experts showed a significant inversion effect for both faces and dogs. In other expert object recognition studies, inversion impairs the speed and accuracy of expert recognition processes (Ashworth, Vuong, Rossion, & Tarr, 2008; Campbell & Tanaka, 2018; Rossion & Curran, 2010; Rossion, Gauthier, Goffaux, Tarr, & Crommelinck, 2016) and limits the visual short-term memory capacity of the expert (Curby, Glazek, & Gauthier, 2009).
Although it has been speculated that mammogram expertise involves holistic strategies (Kundel et al., 2007), direct tests of holistic processing strategies in radiology have yet to be conducted. To investigate a possible link between holistic perception and mammogram expertise, we tested the effects of inversion on a group of experienced radiologists (> 5 years of radiology experience) and radiology residents (< 5 years of radiology experience). On average, an experienced mammographer evaluates between 1000 and 15,000 images per year (Evans et al., 2013) compared to resident radiologists, who see fewer than 300 cases during the course of their clinical training. Expert mammographers and residents have likely received similar formal mammography training, but it is the experts, with their extended experience, who exhibit evidence of rapid detection (Evans et al., 2013; Kundel & Nodine, 1975).
In our study, participants made a “normal/abnormal” decision to briefly presented upright and inverted mammograms. In order to rule out any age-related inversion effect, participants were asked to judge the facial expressions (e.g., neutral/happy) of briefly presented upright and inverted faces. Recognition of facial expressions, like facial identity, recruits holistic perception that is disrupted by inversion (Calder & Jansen, 2005). We made three predictions: First, given that virtually everyone is an expert in holistic expression perception, we expected that both the experienced and resident radiologists would show an inversion effect in their perception of expression (i.e., better detection of happy expressions in upright faces than inverted faces). Second, we hypothesized that the experienced radiologists (< 5 years of radiology practice) would be more accurate in their discriminations of upright mammograms than novice radiology residents. Finally, as evidence of their holistic strategies, we predicted that the experienced radiologists should show a greater inversion effect to mammograms (i.e., difference between upright and inverted recognition) than the resident radiologists.