Awake fMRI reveals a specialized region in dog temporal cortex for face processing

Recent behavioral evidence suggests that dogs, like humans and monkeys, are capable of visual face recognition. But do dogs also exhibit specialized cortical face regions similar to humans and monkeys? Using functional magnetic resonance imaging (fMRI) in six dogs trained to remain motionless during scanning without restraint or sedation, we found a region in the canine temporal lobe that responded significantly more to movies of human faces than to movies of everyday objects. Next, using a new stimulus set to investigate face selectivity in this predefined candidate dog face area, we found that this region responded similarly to images of human faces and dog faces, yet significantly more to both human and dog faces than to images of objects. Such face selectivity was not found in dog primary visual cortex. Taken together, these findings: (1) provide the first evidence for a face-selective region in the temporal cortex of dogs, which cannot be explained by simple low-level visual feature extraction; (2) reveal that neural machinery dedicated to face processing is not unique to primates; and (3) may help explain dogs’ exquisite sensitivity to human social cues.


120
Functional scans used a single-shot echo-planar imaging (EPI) sequence (24 slices, TR=1400 ms, 121 TE=28 ms, voxel size, 3 x 3 x 3 mm, flip angle=70 degrees, 10% gap). Slices were oriented dorsally to the 122 dog's brain (coronal to the magnet, as, in the sphinx position, the dogs' heads were positioned 90 123 degrees from the usual human orientation) ( Fig. 1) with the phase-encoding direction right-to-left. 124 Sequential slices were used to minimize between-plane offsets from participant movement, and the 125 10% slice gap minimized the crosstalk that can occur with sequential scan sequences. We have 126 previously found that both structural and functional resolutions were adequate for localizing activations 127 to structures like the caudate nucleus (Berns et al., 2012;Berns et al., 2013;Cook et al., 2014). 128 Stimuli were presented using Python 2.7.9 and the Expyriment library. Each stimulus block was 129 manually triggered by an observer at the rear of the magnet. This manual triggering ensured that the 130 dog was properly stationed at the beginning of each block. Importantly, no actual human was in view 131 during any of the stimulus presentation blocks. The center of each stimulus was presented binocularly, 132 and at eye level in front of the dog, such that each stimulus fell in the center of the visual field when the 133 dog was looking forward.

134
Heart rate and respiration rate were not collected during scanning. However, we do not believe 135 that either of these physiological measures contributed to our findings for two reasons. First, although 136 dogs' heart rate (HR) and respiration rate (RR) are greater than humans', they are not that much faster. 137 In fact, large dogs (which most of our subjects are) have HRs and RRs similar to humans. Additionally, 138 the TR in this study was 1400 ms, which is about 30% faster than the TR in a typical human study (2000 139 ms). Thus, any modestly faster HR and RR is compensated for by the faster TR. Second, both HR and RR 140 would produce a general effect across brain regions, and we see differential effects between DFA and 141 V1 (see Functional Data Preprocessing and Analysis Section and Results Section).

144
Preprocessing was conducted using AFNI (NIH) and its associated functions, and steps were 145 identical to those described previously (Berns et al., 2012;Berns et al., 2013). In brief, 2-pass, 6-146 parameter affine motion correction was used with a hand-selected reference volume for each dog. We 147 hand selected a reference volume because the first volumes are never the most representative of the 148 dog's head position during the study. The reference volume was typically midway in the first run, after 149 the dog has settled into a comfortable position. Next, because dogs moved between blocks (and when 150 rewarded), aggressive censoring was carried out. A volume was flagged for censoring based on two 152 intensity. Censored files were inspected visually to be certain that bad volumes (e.g., when the dog's 153 head was out of the scanner) were not included. The majority of censored volumes followed the 154 consumption of food. If less than 33% of the volumes were retained, we excluded that subject (Berns et    The average timeseries during stimulus blocks for faces and objects was extracted from the DFA 182 for each dog. Each timeseries was detrended, and values from censored volumes replaced with NaNs. 248 Our fMRI results build on these ERP findings and offer strong evidence for a face-selective region in dog 249 temporal cortex, responding significantly more to images of faces than to images of objects.
250 Furthermore, the face selectivity of the DFA was not found in dog primary visual cortex, ruling out 251 simple low-level feature extraction as explanations for the face-selective response in DFA.

252
The principal limitation of our study stems from the small effect size. The average differential 253 BOLD response was well less than 1%, which is consistent with human fMRI studies. Comparable animal 254 fMRI studies, however, overcome the signal limitation by immobilizing the subject and scanning for 255 much longer periods of time to decrease the effects of noise, but this approach often uses a small 256 number of subjects -typically two. In contrast, our approach is to use awake, trained dogs who 257 cooperatively enter the scanner and hold still for periods up to several minutes without restraint. And 258 while the dogs do extraordinarily well, the data quality cannot approach that obtained from a sedated, 259 immobilized monkey. Thus, the trade-off is noisier data. We compensate by using more subjects than a 260 typical monkey study, here reporting the data from six dogs. Although we have studied larger cohorts of 261 dogs in previous studies, watching images on a flat screen is not a natural behavior for dogs, and only a 262 subset of the MRI-trained dogs would do so, even after months of training. Even so, the data we report 263 here show a high degree of within-subject replicability, with some inter-subject variation in the location 264 of DFA, some of which may be due to noise and some due to the existence of multiple face-sensitive 265 patches. Another potential limitation of our study may be the concern about vasculature effects. In 266 fMRI, signal changes in a given region of cortex are attributed to neuronal activity. However, it could be 267 the case that such fMRI signal changes might arise from a draining vein, making it difficult to say 268 whether the fMRI signal changes are due to neuronal activity in that region of cortex, in more distant 269 cortical regions, or both. Physiological noise may hypothetically affect the detected activations; 270 however, we have no a priori reason to suspect that the reported DFA or V1 activities are due to 271 physiological confounds. The lack of condition-specific effects in V1 rules out a global confound, so the 272 remaining question is whether the putative DFA activity is a result of physiologic noise on a local level. In 273 a previous dog-fMRI study (Cook et al., 2014), we investigated the inclusion of a ventricle ROI as a 274 covariate and proxy for physiological noise and found that it was not a significant contributor to 275 activations in the reward system. In that experiment, the stimuli represented conditioned signals to food 276 reward, and would be expected to be far more arousing than the visual stimuli used here. Thus, the 277 ultimate goal is to obtain converging evidence across multiple methodologies (fMRI, neurophysiology, 278 lesion studies, etc.) and across multiple labs to definitively establish the selectivity of a given cortical 1 Experimental setup in MRI.
Dogs were trained to station within an individually customized chin rest placed inside a stock human neck coil. The upper surface coil was located just superior to the dog's head. Images were rear projected onto a translucent screen placed at the end of the magnet bore. In the dynamic stimuli runs, color movies clips (3-s each) were shown in 21 s blocks of human faces, objects (toys), scenes, and scrambled objects. In the static stimuli runs, black and white images (600 ms on, 400 ms off) were shown in 20 s blocks of human faces, dog faces, everyday objects, scenes, and scrambled faces. The dynamic stimuli runs were used to localize a candidate face region in the temporal cortex of dogs, and then the static stimuli runs were used to independently test the face selectivity of this region.