Virtual brain reading: A neural network approach to understanding fMRI patterns.
Is the fusiform face area (FFA) a module specialized for processing faces, or does it simply support generic visual expertise? Researchers have investigated this question using Multi-Voxel Pattern Analysis (MVPA) applied to fMRI results. Haxby et al. (2001) showed that patterns of neural activation in object-selective visual cortex can be used to discriminate object categories, even when voxels selective for those categories are removed. This provided evidence for a distributed neural code, in which information about faces exists outside the FFA. In contrast, Spiridon and Kanwisher (2002) showed that activation patterns seen in face-selective cortex were more effective for making face vs. non-face discriminations than for non-face vs. non-face discriminations, whereas this was not true for other object categories, such as man-made objects. This implied FFA neurons contain special information about faces, but that there is no specialized module for other object categories.
We present a neurocomputational model of visual processing, in which object representations are organized topographically. Photographic images are subject to Gabor filtering, then PCA, then input to a Kohonen network - a self-organizing neural network that groups similar inputs together, forming a two-dimensional “semantic map” of stimulus space. We present a new method for “virtual MVPA”, in which we assume that activations of units in the Kohonen network layer correspond to neural activity in ventral visual cortex, and may be mapped onto voxel activations measured by fMRI. We trained the model on images of cups, cans, books and faces. The Kohonen network developed a region dedicated to each category. In line with Haxby et al. (2001), each of these dedicated regions still had different response patterns to stimuli from the other categories, such that activation patterns in areas of the semantic map dedicated to one category (e.g., faces) can be used to distinguish between other categories (e.g., cups versus cans). However, in line with Spiridon and Kanwisher, the face area is better at distinguishing faces from non-faces than at distinguishing non-face categories from each other, while non-face areas of the semantic map are, on average, equipotent at both tasks. In the model, this can be explained by lower within-category variability of the representations of faces compared to, say, cups. This is due to higher within-category visual similarity for faces than for other categories. Hence, with a model of visual cortex possessing no special mechanism for face processing, we simulate Spiridon and Kanwisher’s results, casting doubt on their interpretation in favor of a specialized face module.
Conference:
Computational and systems
neuroscience 2009, Salt Lake City, UT, United States, 26 Feb - 3 Mar, 2009.
Presentation Type:
Poster Presentation
Topic:
Poster Presentations
Citation:
(2009). Virtual brain reading: A neural network approach to understanding fMRI patterns..
Front. Syst. Neurosci.
Conference Abstract:
Computational and systems
neuroscience 2009.
doi: 10.3389/conf.neuro.06.2009.03.338
Copyright:
The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers.
They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.
The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.
Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.
For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.
Received:
04 Feb 2009;
Published Online:
04 Feb 2009.