Consider how many people you can recognize, including friends, family, celebrities, or just a face you associate with a certain event or memory. Prosopagnosia, or the inability to recognize faces, is a condition caused by damage to brain regions used for facial recognition. The exact processes involved in facial coding are not completely understood; a recent article in Cell explores a new concept in the field known as Axis Coding.
The areas most closely associated with primate and human facial coding are located in the inferior temporal cortex (ITC). Researchers focused on 2 specific groups of neurons within the ITC, the middle lateral (ML)/middle fundus (MF) and the anterior medial (AM) regions. Electrophysiological recordings were taken from a total of 205 face associated cells (106 ML/MF and 99 AM) in the ITCs of 2 macaque monkeys. Each monkey was exposed to 2,000 different faces, generated to contain 50 dimensions that could be recognized by the face cells. The subject’s were exposed to these stimuli at least 3-5 times before all the data was pooled and analyzed using various statistical methods.
The results contradicted previous studies that explain facial coding through an Exemplar Model which posits that specific faces will elicit specific responses from cellular regions in the ITC. Simply put, the face of your grandmother would always excite the same cells the same way because that is how this familiar stimulus has been encoded. This study countered that model with an Axis Coding Model which postulates that certain aspects of faces are encoded based on their position on a specific dimensional axis and that the ML/MF and AM regions encode complimentary information about these facial aspects. Further, it also predicts that certain planes orthogonal (at a right angle) to the dimension recognized by the cell would exhibit a specific frequency of firing. This was confirmed in further analysis and is what makes this model more flexible and powerful than the Exemplar Model. Most interestingly, researchers were able to reconstruct certain faces using only the neuronal recordings collected. This means that using the output of only 205 cells in the ITC, they were able to essentially guess, or decode, which face the subjects were looking at with significant accuracy.
This study used an innovative approach to discover a new way of investigating a long studied aspect of facial coding. These researchers were also able to streamline the complex process of facial recognition to a simple linear model of encoding and decoding and represents a new avenue of understanding for those with damage to areas implicated in this process. Of course, more research must be conducted in human models to most efficiently apply this knowledge in a clinical setting.
Written By: Clifton Lewis