Projects
Predicting brain function from anatomy using geometric deep learning
Retinotopic mapping in human visual cortex is known to be remarkably similar across participants; however, considerable inter-individual variation does exist, and this variation has been shown to be directly related to variability in cortical folding patterns and other anatomical features. In fact, previous studies have shown that one can quite reasonably predict the functional organization of the visual cortex in participants by warping an atlas or template (Benson et al., 2014, 2012; Benson and Winawer, 2018) according to their specific anatomical data. While these approaches are able to provide reasonable estimates of their retinotopic maps, they have not – when using anatomical information alone – been able to capture the detailed idiosyncrasies seen in the actual measured maps of those participants. Thus, we aimed to build a neural network able to predict individual differences in retinotopic maps by solely exploiting differences in anatomical features of the cortical surface.
Retinotopic maps are perhaps best represented using cortical surface models, which is possible as most of the retinotopically organized areas are located in the cerebral cortex (i.e., the outermost tissue layer of the brain). By representing the data on a cortical surface model, it is possible to visualize the retinotopic maps in such a way that the detailed functional organization of each specific area can be appreciated as well how the layout of each visual area relates to one another more globally. It has hence become common place for retinotopy data to be represented in such a way. Pertinently, the Human Connectome Project (Van Essen et al., 2013) 7T retinotopy dataset (Benson et al., 2018), the largest publically available high-resolution retinotopic mapping dataset, uses a cortical surface representation for all its retinotopy data. Here, we sought to leverage geometric deep learning techniques (Bronstein et al., 2017) to build a neural network able to interact directly with the cortical surface representations of this large open-source HCP dataset to predict individual-specific retinotopic maps.
Related conferences
