Large-Scale Multimodal Brain Network Construction and Its Application to Maltreated Children

• Activity Number: 89
• Type: Invited
• Date/Time: Sunday, August 9, 2015 : 8:30 PM to 9:15 PM
• Sponsor: Korean International Statistical Society
• Author(s): Moo Chung*
• Companies: University of Wisconsin - Madison
• Keywords: Large-scale brain networks ; persistent homology ; topological data analysis ; small-n large-p problem ; MRI ; DTI

Abstract

We explore the computational issues of constructing large-scale whole brain networks with more than 100000 nodes by taking each voxel as a network node in 3D brain images. Existing brain network models exhibit serious defects in applying to the whole brain regions due to the small-n large-p problem. Specifically, the number of nodes p is substantially larger than the number of subjects n so the often used sample covariance and correlation matrices show the rank deficiency and they are no longer positive definite. We propose to remedy the shortcoming of the small-n large-p problem by persistent homology, a recently popular topological data analysis framework. By exploiting the hidden but persistent topological structures in the sparse network construction, the computation that takes more than 56 hours using existing methods can take mere 10 seconds. The method is applied to MRI and DTI of maltreated children. We show that the maltreated children exhibit consistent white matter abnormality across different imaging modalities. This study is based on http://arxiv.org/abs/1409.0177.