Data-driven modeling of BOLD drug response curves using Gaussian process learning

Research output: Chapter in Book/Report/Conference proceedingConference paper

2 Citations (Scopus)

Abstract

This paper presents a data-driven approach for modeling the temporal profile of pharmacological magnetic resonance imaging (phMRI) data, in which the blood oxygen level-dependent (BOLD) response to an acute drug challenge is measured. To date, this type of data have typically been analysed using general linear models applied to each voxel individually, an approach that requires a pre-defined model of the expected response to the pharmacological stimulus. Previous approaches have defined this model using pharmacokinetic profiles, phMRI data from pilot studies, cognitive or physiological variables that have been acquired during the experiment or a simple pre-post boxcar profile. In contrast, the approach presented here is data-driven; a basis function is fitted to the data in a Bayesian framework using Gaussian processes. This method outperforms two previous multivariate approaches to fMRI analysis while also providing information about the shape of the BOLD response and hence, increasing the model interpretability.
Original languageEnglish
Title of host publicationMachine Learning and Interpretation in Neuroimaging
Subtitle of host publicationInternational Workshop, MLINI 2011, Held at NIPS 2011, Sierra Nevada, Spain, December 16-17, 2011, Revised Selected and Invited Contributions
PublisherSpringer
Pages210-217
Number of pages8
ISBN (Electronic)978-3-642-34713-9
ISBN (Print)978-3-642-34712-2
DOIs
Publication statusPublished - 2011
EventNeural Information Processing Systems - Grenada, Spain
Duration: 10 Nov 2011 → …

Publication series

NameLecture Notes in Computer Science
Volume7263

Conference

ConferenceNeural Information Processing Systems
Country/TerritorySpain
CityGrenada
Period10/11/2011 → …

Fingerprint

Dive into the research topics of 'Data-driven modeling of BOLD drug response curves using Gaussian process learning'. Together they form a unique fingerprint.

Cite this