|
Abstract:
|
The human cerebral cortex is a folded thin layer of grey matter. It is now well established that it is beneficial to parametrize this thin layer as 2D surface rather then a 3D volume. In this context classical individual analysis techniques, as the Seed-Based Correlation Analysis, can be applied to derive Functional Connectivity Maps associated to the cerebral activity over the cortical surface. However, when it comes to multi-subject studies of these maps, classical tools as Principal Component Analysis fail to give interpretable results, mostly due to the high dimensionality of the data. We propose the application of a novel Regularized Principal Components algorithm as a dimensionality reduction tool that allows the study of the main modes of variation of the Functional Connectivity Maps. This algorithm naturally deals with data lying on a 2D surface thanks to the introduction of a smoothness penalty in the model. The smoothing penalty considered is coherent with the 2D geodesic distances over the cortical surface. We consider the algorithm in conjunction with data from the Human Connectome Project and investigate connectivity maps and their variation.
|
