Technical Program

Paper Detail

Paper: PS-1B.18
Session: Poster Session 1B
Location: H Fl├Ąche 1.OG
Session Time: Saturday, September 14, 16:30 - 19:30
Presentation Time:Saturday, September 14, 16:30 - 19:30
Presentation: Poster
Publication: 2019 Conference on Cognitive Computational Neuroscience, 13-16 September 2019, Berlin, Germany
Paper Title: Using deep neural network features to predict voxelwise activity in ultra-high field fMRI
Manuscript:  Click here to view manuscript
License: Creative Commons License
This work is licensed under a Creative Commons Attribution 3.0 Unported License.
DOI: https://doi.org/10.32470/CCN.2019.1165-0
Authors: Rebekka Heinen, Lorena Deuker, Ruhr University Bochum, Germany; Thomas Naselaris, Medical University of South Carolina, United States; Nikolai Axmacher, Ruhr University Bochum, Germany
Abstract: Deep neural network features can be used to train encoding models that accurately predict brain activity from the visual cortex. Using these features together with ultra-high field fMRI could open a new set of opportunities ranging from human vision to areas such as learning and memory consolidation. Is it possible to apply encoding models based on deep neural network features to high-resolution fMRI data? We investigated this using the feature-weighted receptive field (fwrf) model on ultra-high field fMRI during a natural image viewing task. Applying the fwrf model to our data we were able to predict brain activity along the ventral visual stream (VVS). In line with previous studies, we found a shift from low to high network layers while predicting brain activity in early visual areas compared to higher regions of the VVS. We conclude that encoding models based on neural network features can be applied to ultra-high field fMRI data, suggesting similar processing of visual scenes in neural networks and the human visual association cortex. Our results suggest that these models cannot only be used to study vision but other processes such as memory and imagination.