English
 
Help Privacy Policy Disclaimer
  Advanced SearchBrowse

Item

ITEM ACTIONSEXPORT

Released

Meeting Abstract

Task fMRI Prediction from Task Free Resting State Measurements

MPS-Authors
/persons/resource/persons214742

Lacosse,  E
Department High-Field Magnetic Resonance, Max Planck Institute for Biological Cybernetics, Max Planck Society;
Max Planck Institute for Biological Cybernetics, Max Planck Society;

/persons/resource/persons133483

Lohmann,  G
Department High-Field Magnetic Resonance, Max Planck Institute for Biological Cybernetics, Max Planck Society;
Max Planck Institute for Biological Cybernetics, Max Planck Society;

/persons/resource/persons216033

Himmelbach,  M
Department High-Field Magnetic Resonance, Max Planck Institute for Biological Cybernetics, Max Planck Society;
Max Planck Institute for Biological Cybernetics, Max Planck Society;

/persons/resource/persons84187

Scheffler,  K
Department High-Field Magnetic Resonance, Max Planck Institute for Biological Cybernetics, Max Planck Society;
Max Planck Institute for Biological Cybernetics, Max Planck Society;

Fulltext (public)
There are no public fulltexts stored in PuRe
Supplementary Material (public)
There is no public supplementary material available
Citation

Lacosse, E., Lohmann, G., Himmelbach, M., Scheffler, K., & Martius, G. (2018). Task fMRI Prediction from Task Free Resting State Measurements. In 11th International Conference on Brain Informatics (BI 2018) (pp. 47).


Cite as: http://hdl.handle.net/21.11116/0000-0002-A92C-F
Abstract
BOLD fMRI is widely used to study taskinduced signal changes (tfMRI) in conjunction with taskfree, i.e., restingstate (rsfMRI). However, an understanding of how these two types of measurements when utilized together can robustly infer principles of human brain organization and behaviour is still lacking. Specifically, previous literature sought to understand connections between rsfMRI and tfMRI measurements by demonstrating that functional correlation properties during rsfMRI remain largely unchanged during cognitive tasks. Here, we reevaluate recent work building on this link. The core idea of those works examined models with the specific goal of regressing features derived from rsfMRI scans onto tfMRI general linear model (GLM) maps to predict interindividual differencesa vital target for increasing the usefulness of fMRI in clinical settings. First, we introduce a baseline model evaluation metric for empirical results previously reported in the literature and reproduce results next to this metric. This baseline evaluation provides an essential component under which we interpret quantitative results obtained with machine learning methods used in this context. Second, we show how observing a dependence between tfMRI and rsfMRI may arise under different encoding models. This demonstration emphasizes (a) the need for proper model evaluation and (b) the need to thoroughly examine the implications of observing higher withinsubject prediction scores than betweensubject scores. Third, we show how improvements in previously established regression methods can be made using a regularization technique. We demonstrate this with empirical results on the latest data from the Human Connectome Project (HCP) release. Although it is known that intrinsic differences in rsfMRI and tfMRI patterns can be characterized to a certain extent, our work stresses the need for proper model evaluation in reporting quantitative results. Taken together, our work provides a new lens for the evaluation of previously reported results and an important commentary to a topic that has generated a great deal of attention and interest.