A prototypical template for rapid face detection is embedded in the monkey superior colliculus

概要

Background. Face information processing consists of several different steps; 1) face detection based on first order information (i.e., arrangement of facial features such as eyes, nose, mouth, etc.), 2) holistic processing of facial features (integration of facial features into a whole-face representation), and 3) facial discrimination based on second-order information (discrimination of variance across faces). Although computational studies suggest that the first face detection step precedes the others and is important to make the later processes efficient, neural mechanisms of this step remain poorly understood. Our previous neurophysiological studies reported that monkey superior colliculus (SC) neurons responded to facial photos and their latencies were shorter than those in the striate cortex. These findings suggest that the SC is involved in face detection. However, whether SC neurons respond to first order information of faces (basic layout of facial features including eyes, nose, and mouth) itself or not remains unknown. To test this hypothesis, we recorded the responses of monkey SC neurons to face-like and non-face patterns, in which facial features were either orderly or randomly positioned within contours. Furthermore, the responses to inverted (upside down) and contrast-reversed patterns of the same face-like and non-face patterns were examined as well. We looked for neurons coding a face template in the SC which would respond stronger to both normal and contrast- reversed face-like patterns as well as upright and inverted face-like patterns than non-face patterns.

Methods. Two adult (1 female and 1 male) macaque monkeys (Macaca fuscata) were used. The monkeys were trained to perform a sequential delayed non-matching-to-sample task (DNMS) that required the discrimination of the visual stimuli. These stimuli consisted of two different sets of stimuli [white (normal) and black (contrast-reversed) stimulus sets]. Each stimulus set included 4 stimulus groups with 4 different contours (rice scoop, star, circle, square). Each stimulus group included 5 visual stimuli (forms) that had the same contour; 2 face-like patterns (upright, inverted), and 3 non-face patterns [2 random patterns (random1, random2) and blank contour only (blanks)]. Each face-like pattern consisted of one of 4 face contours and 5 facial features (2 eyes, 2 eyebrows, 1 mouth), while the random patterns consisted of the same facial contour and features as in the face-like patterns, but the facial features were randomly positioned within the facial contour.

Neuronal activity was recorded from the monkey SC during the DNMS task. Response magnitudes and latencies to these stimuli were analyzed. To analyze representation of the stimulus by population activity of the SC neurons, activity of the SC neurons was analyzed by multidimensional scaling (MDS).

Results. Of 158 visually responsive SC neurons, 146 neurons were tested with all of the visual stimuli. Based on the response areas in the visual field (VF), 146 neurons were divided into three groups: neurons with the largest response areas in the upper VF [dubbed “SC neurons with upper receptive fields (RFs)”; n = 46], neurons in the lower VF (68 with lower RFs), and those in the central VF (32 with central RFs). Statistical analyses of the neuronal responses to the stimuli indicated that SC neurons responded stronger and faster to the upright and inverted face-like patterns compared to the nonface patterns, regardless of contrast polarity and contour shapes. Furthermore, SC neurons with central receptive fields were more selective to face-like patterns. In addition, the population activity of SC neurons with central receptive fields discriminated the face-like patterns from the nonface patterns as early as 50 ms after the stimulus onset. The results provide strong neurophysiological evidence for the involvement of the primate SC in face detection and suggest the existence of a broadly tuned template for face detection in the subcortical visual pathway including the SC.

Conclusions. The present evidence provides a neurophysiological basis for an idea that the subcortical visual pathway including the SC in vertebrates functions as a prototypical face template, i.e., “Conspec” suggested in behavioral studies. Consistently, the SC neurons responded best to the face-like patterns regardless of 4 different contours as well as regardless of contrast polarity of the stimuli in the present study.