Meeting Banner
Abstract #2204

Network Analysis Based on Analytic Solution to Permutation Tests on Support Vector Machines

Madhura Ingalhalikar1, Bilwaj Gaonkar1, Alex R. Smith1, Robert T. Schultz2, Ragini Verma1

1Section of Biomedical Image Analysis, Department of Radiology, University of Pennsylvania, Philadelphia, PA, United States; 2Center for Autism Research, Childrens Hospital of Philadelphia, Philadelphia, PA, United States

The study employs a novel method using an analytic approach to permutation tests on support vector machines for computing statistical significance maps on connectivity matrices. Permutation tests are critical for interpreting SVM output for high dimensional data. However performing these tests is time consuming and computationally expensive. However, the analytical approximation to these tests can yield the results quickly. We apply this method to investigate the differences between patients with autism and typically developing controls based on their structural connectivity networks. We find that patients with autism show lower connectivity mainly in the connections initiating from temporal regions like fusiform gyrus and insula as well between fronto-parietal tracts.

Abstract PDF

Keywords

actual ados affected affine allow analyses analytic analytical analyze applied applying approximation aside autism biomedical blue bottom brain channel characterization classifiers clinical coefficients coil computation computational computationally computing conditional connecting connection connections connectivity consensus consisted controlling controls cortical dataset deficit dependence derived developing diagnosed diagnosis differ disorders displayed edge edges employs features fibers fraction functional generally global gold gradient grant grants graph harnesses head hemisphere identify identifying implies include inter intra involve involving just known labels linear local machines male maps mass matched matrices memory modeled moreover multivariate network networks nodes normalized novel objective orange pathway pathways permutation plagues potential preprocessing primary probabilistic problem processes proposed psychologists radiology reaching recent reduced regression replaces represent role section segmentation shot significance significantly smith social solution solve space spectrum starting statistic statistical structural structures studies subject subjects suggesting support supported table takes target temporal thereby topological transferred typically univariate vector view visualization whole wise