Neural Mechanism of Spatial Frequency Representation in Face Categorization: an ECoG Study*

Abstract

Converging evidences show that face categorization and recognition are represented in specific brain areas. It is also known from behavioral studies that spatial frequency of pictures takes different roles in different face categorizations, e.g. identity information is carried more by low frequency features, and gender is represented by both high and low frequencies while high frequency is more important for face expression. However, there is no solid experimental data about the neural mechanism of the different contributions of spatial frequencies on face categorization up to now. ECoG was collected from epilepsy patients while they underwent a monitoring session to locate their epilepsy foci with implanted subdural surface electrodes. Pictures with different gender, facial expression and identities were presented to the patients, while they were required to perform a simple detection task to maintain their focus on the pictures. The changes of traditional face related components, N170, were analyzed by event related potentials and the significant changes was also verified on each electrode by permutation test. The N170 latency is found to be delayed for high spatial frequency (HSF) pictures. For faces of unknown person, low spatial frequency (LSF) presentation results in a longer N170 latency, but not for faces of well-known movie stars. Female faces show longer N170 latency in HSF conditions compared to LSF, but male faces have no such effect. No significant effect was found on N170 component for expression. However, analysis based on individual electrodes showed that more frontal electrodes are involved in expression representation; the identity-specific sites are more likely to respond to LSF stimuli, and the gender-specific sites have equal responses to both LSF and HSF stimuli, while the expression-specific sites are also more likely to respond to the LSF stimuli, which is inconsistent with existing behavioral studies, and show significant differences as early as 114 milliseconds. It fits with the cognitive model that expression related information has been analyzed briefly in occipital-temporal areas in the early stage before other brain areas are involved for further processing. In summary, spatial frequency's contributions to identity and gender could be represented by N170 in traditional face related brain areas while expression information, instead of carrying by N170, is distributed in a wider area in occipital temporal region for fast analysis and relayed to other areas, e.g. frontal cortex. This is the first study to study the neural mechanism of spatial frequency's contributions in face categorization by ECoG, and provides a new perspective to understand the brain dynamics of feature processes for face perception.