Personal tools

Extended dual regression for account for inidivual variabilites in groupICA

From hpcwiki

Revision as of 09:43, 4 September 2013 by Fgomez (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Contents

People

  • Francisco Gómez

Summary

Independent component analysis (ICA) is one of most popular methods to study fMRI resting state activity. In group studies, a common strategy is to concat individual data along temporal dimension and then looking for maximally independent spatial sources that summarize the functional connectivity at group level, this approach is called group spatial ICA. This group data decomposition is typically followed by an estimation of specific individual information extracted out of these averages maps. The most common method to perform this task is dual regression that consists in two stages: 1) regressing out the average spatial maps into the 4D data subject data to obtain a set of subject-timecourses, followed by a 2) a regresing out of these timecourses into the same 4D dataset to obtain subject specific spatial maps. Following, these individual spatial maps are tipically entered to a second level analysis. Individual extraction information is critical in order to characterize subject specific information. Specially, in conditions where subjects will have subject-specific sources of noise, which will violate the underlaying assumptions of homogeneity behind spatial group ICA. In addition, the usual approach to perform this dual regression based on univariate regression may ignore important information about spatial relationships typically present at subject level. In this work, we propose to extend the dual regression approach for group ICA to account for specific subject information related with spatial distribution and specific noise sources. We expect to test the new method in disorder of consciousness patients where large neighborhoods of spatial relationships has been previously exploited for the signal characterization and where is expected to have large heterogeneity in the individual noise sources.

Method overview

Data sources

Results (Expected)

  • Method conference. Framework proposal.
  • Journal article. Complete framework report.

References