Pain is a subjective experience communicated to others to alert them of potential threats or to seek assistance [1, 2]. Effective communication of pain can operate through verbal and non-verbal cues. Among the non-verbal cues, facial expression is considered one of the most reliable indicators of pain [3, 4, 5]. The effective recognition of facial expressions of pain is of utmost importance, particularly in nonverbal populations such as infants [6, 7] and adults with dementia [8, 9]. However, for its communicative function to be fulfilled, the observer must accurately decode the facial expressions.
Studies investigating how pain is encoded through facial expressions have highlighted the occurrence of three features: the contraction of the eyebrows, the wrinkling of the nose with the raising of the upper lip, and the narrowing of the eyes [10, 11, 12]. The decoding of pain facial expressions relies on the processing of these features [13, 14]. Thus far, the study of low-level visual information underlying this processing has led to inconsistent findings. One of the first steps in vision concerns the decomposition of the visual signal in different spatial frequency (SF) bands [15]. Low SFs (LSFs; see Fig. 1) convey the coarse structures used when viewing faces from a distance [16] or in periphery [17] whereas high SFs (HSFs) convey edges and fine details available when faces are viewed from closer and at the fovea [18]. Two recent articles suggest that LSFs play a central role in the recognition of facial expressions of pain [19, 20]. However, as a communication signal, one would expect that pain signal would be best suited to a short-to-medium distance where one can benefit from immediate assistance. Furthermore, our previous work suggests that SFs between 11 and 21 cycles per face (cpf) are the most useful [14, 21]. According to the terminology used in previous research [19] and explained in more detail below, these SFs would correspond to mid SFs (MSFs).
One potential explanation for this discrepancy could lie with Wang and collaborators’ [19] use of cutoffs to isolate the impact of LSF and HSF. In the literature, the cutoffs used to define LSF and HSF vary arbitrarily (e.g. LSF defined below 8 cpf in [19]; between 2 and 8 cpf in [22]; below 6 cycles per image in [23]; and HSF defined above 32 cpf in [19, 24] or above 24 cycles per image [23]). To the best of our knowledge, such variation is not theoretically driven and is often informed by methodological issues (e.g.[25, 26]) or for replication purposes [27]. Furthermore, the use of such cutoffs hinders the potential contribution of MSFs, leading to an incomplete or incorrect account of the role of SF in pain perception.
The objective of this study is to offer a more complete account of the role of SF in pain perception by incorporating findings from two different methods. Therefore, experiment 1 aimed to reveal which SFs are the most useful for pain categorization among other emotional expressions (i.e. anger, disgust, fear, joy, sadness, surprise, neutral or pain). This kind of experiment is standard in the facial expression literature (e.g. [14, 19]) but instead of using cutoffs to create low-pass and high-pass filters [19], we used SF Bubbles. The fundamental basis of the Bubbles method and its variants (e.g. SF Bubbles, orientation Bubbles), is that it allows the random sampling of information (e.g. local image features, SFs, or orientations, see refs [28–38]) contained in a visual stimulus in order to reveal the relative importance of this information for efficient visual processing. Here in the SF Bubbles method, SFs contained in facial expression images were randomly sampled on each trial (see the Methods section for more details on the stimuli creation procedure), allowing to calculate the probability that participants will accurately identify the facial expression presented based on the presence or absence of certain SFs. Therefore, if the sampled SFs are useful for processing a particular facial expression, it will increase the likelihood that participants will respond accurately, and conversely, if they are not useful, it decreases the likelihood that participants will respond accurately.
Experiment 2 aimed to reveal the optimal SFs for pain recognition through the manipulation of the face retinal size (equivalent to the distance between the stimulus and the observer). As in experiment 1, participants were asked to categorize the perceived facial expressions as corresponding to anger, disgust, fear, joy, sadness, surprise, neutral or pain although face images were presented in different sizes. The objective of this experiment was to verify the impact of distance on the ability to categorize the facial expression of pain. Since layers of HSFs are progressively peeled off by increasing distance between the observer and the distal stimulus, this experimental manipulation also allows to investigate the role of relatively high SFs in the presence of lower SFs. This method is also considered more ecological since in everyday life, the distance at which one sees people’s facial expressions can vary considerably.