Purchase this article with an account.
Mark Vida, Marlene Behrmann; Visualizing the Spatiotemporal Dynamics of Neural Representations of Individual Face Identities. Journal of Vision 2015;15(12):201. doi: https://doi.org/10.1167/15.12.201.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Individual face identities are represented in the human brain by spatially distributed patterns of neural activity (e.g., Nestor et al., 2011). We investigated the spatiotemporal dynamics of this representation. An adult human viewed well-controlled color photographs of 91 male face identities (two expressions and 104 trials per identity), while brain activity was recorded with magnetoencephalography (MEG). For 47 time points 10-470 ms after stimulus onset, and two regions-of-interest (ROIs) commonly implicated in face processing (right and left posterior fusiform gyrus [pFG]), we used linear SVM classification to measure neural discrimination of all possible face pairs, across facial expression. We then used multidimensional scaling to identify statistical dimensions underlying the neural discrimination data. For each dimension and time point, we constructed a “classification image” (CI), which displays visual information captured by the dimension. CIs were constructed by computing the difference between weighted averages of faces on each side of a given dimension, and then applying a permutation test to isolate statistically significant pixels. Across all face pairs, left and right pFG displayed the best classification performance at 180-200 ms. For both ROIs, the proportion of significant pixels in the CIs was highest at around 200 ms, with smaller peaks at around 100 and 350 ms. Hence, information about face identity was captured most clearly at around 200 ms. Pixel-wise correlations between CIs for the two ROIs were small to moderate (r range = -.03-.28), with correlations peaking at around 200 ms. This pattern suggests that left and right pFG carry partially overlapping information about face identity, with the greatest overlap at around 200 ms. Together, these results suggest that visual information about face identity is represented maximally, but not exclusively, at around 200 ms in pFG, and that the spatial properties of the information represented differ between left and right pFG.
Meeting abstract presented at VSS 2015
This PDF is available to Subscribers Only