Independent component analysis applied to fMRI data: A generative model for validating results

Vince Daniel Calhoun, Godfrey Pearlson, Tulay Adali

Research output: Contribution to journalArticle

Abstract

Methods for testing and validating independent component analysis (ICA) results in fMRI are growing in importance as the popularity of this model for studying brain function increases. We introduce and apply a synthesis/analysis model for analyzing functional Magnetic Resonance Imaging (fMRI) data using ICA. Classes of signal types relevant to fMRI are described and a statistical approach for validation of simulation results is developed. Additionally, we propose an empirical version of our validation approach to test the performance of various ICA approaches in "hybrid" fMRI data, a mixture of real fMRI data and known (validatable) sources. The synthesis portion of the model assumes statistically independent spatial sources in the brain. We also assume that the fMRI scanner acquires overdetermined data such that there are more time points than hemodynamic brain sources. We propose several signal classes relevant to fMRI and discuss the properties of each. The analysis portion of the model includes several candidates for spatial smoothing, ICA algorithm, and data reduction. We use the Kullback-Leibler divergence between the estimated source distributions and the "true" distributions as a measure of the optimality of the final ICA decomposition. Using this model, we generate fMRI-like data and optimize the analysis stage as a function of ICA algorithm, data reduction scheme, and spatial smoothing. An example of how our synthesis/analysis model can be used in validating an fMRI experiment is demonstrated using simulations and "hybrid" fMRI data.

Original languageEnglish (US)
Pages (from-to)281-291
Number of pages11
JournalJournal of VLSI Signal Processing Systems for Signal, Image, and Video Technology
Volume37
Issue number2-3
DOIs
StatePublished - Jun 2004
Externally publishedYes

Fingerprint

Generative Models
Functional Magnetic Resonance Imaging
Independent component analysis
Independent Component Analysis
Brain
Data Reduction
Synthesis
Model Analysis
Smoothing
Data reduction
Magnetic Resonance Imaging
Kullback-Leibler Divergence
Hemodynamics
Scanner
Model
Optimality
Simulation
Optimise
Decomposition
Decompose

Keywords

  • Brain
  • fMRI
  • Independent component analysis
  • Model

ASJC Scopus subject areas

  • Electrical and Electronic Engineering
  • Information Systems
  • Signal Processing

Cite this

Independent component analysis applied to fMRI data : A generative model for validating results. / Calhoun, Vince Daniel; Pearlson, Godfrey; Adali, Tulay.

In: Journal of VLSI Signal Processing Systems for Signal, Image, and Video Technology, Vol. 37, No. 2-3, 06.2004, p. 281-291.

Research output: Contribution to journalArticle

@article{0f142410360746f9a1998d745950643d,
title = "Independent component analysis applied to fMRI data: A generative model for validating results",
abstract = "Methods for testing and validating independent component analysis (ICA) results in fMRI are growing in importance as the popularity of this model for studying brain function increases. We introduce and apply a synthesis/analysis model for analyzing functional Magnetic Resonance Imaging (fMRI) data using ICA. Classes of signal types relevant to fMRI are described and a statistical approach for validation of simulation results is developed. Additionally, we propose an empirical version of our validation approach to test the performance of various ICA approaches in {"}hybrid{"} fMRI data, a mixture of real fMRI data and known (validatable) sources. The synthesis portion of the model assumes statistically independent spatial sources in the brain. We also assume that the fMRI scanner acquires overdetermined data such that there are more time points than hemodynamic brain sources. We propose several signal classes relevant to fMRI and discuss the properties of each. The analysis portion of the model includes several candidates for spatial smoothing, ICA algorithm, and data reduction. We use the Kullback-Leibler divergence between the estimated source distributions and the {"}true{"} distributions as a measure of the optimality of the final ICA decomposition. Using this model, we generate fMRI-like data and optimize the analysis stage as a function of ICA algorithm, data reduction scheme, and spatial smoothing. An example of how our synthesis/analysis model can be used in validating an fMRI experiment is demonstrated using simulations and {"}hybrid{"} fMRI data.",
keywords = "Brain, fMRI, Independent component analysis, Model",
author = "Calhoun, {Vince Daniel} and Godfrey Pearlson and Tulay Adali",
year = "2004",
month = "6",
doi = "10.1023/B:VLSI.0000027491.81326.7a",
language = "English (US)",
volume = "37",
pages = "281--291",
journal = "Journal of Signal Processing Systems",
issn = "1939-8018",
publisher = "Springer New York",
number = "2-3",

}

TY - JOUR

T1 - Independent component analysis applied to fMRI data

T2 - A generative model for validating results

AU - Calhoun, Vince Daniel

AU - Pearlson, Godfrey

AU - Adali, Tulay

PY - 2004/6

Y1 - 2004/6

N2 - Methods for testing and validating independent component analysis (ICA) results in fMRI are growing in importance as the popularity of this model for studying brain function increases. We introduce and apply a synthesis/analysis model for analyzing functional Magnetic Resonance Imaging (fMRI) data using ICA. Classes of signal types relevant to fMRI are described and a statistical approach for validation of simulation results is developed. Additionally, we propose an empirical version of our validation approach to test the performance of various ICA approaches in "hybrid" fMRI data, a mixture of real fMRI data and known (validatable) sources. The synthesis portion of the model assumes statistically independent spatial sources in the brain. We also assume that the fMRI scanner acquires overdetermined data such that there are more time points than hemodynamic brain sources. We propose several signal classes relevant to fMRI and discuss the properties of each. The analysis portion of the model includes several candidates for spatial smoothing, ICA algorithm, and data reduction. We use the Kullback-Leibler divergence between the estimated source distributions and the "true" distributions as a measure of the optimality of the final ICA decomposition. Using this model, we generate fMRI-like data and optimize the analysis stage as a function of ICA algorithm, data reduction scheme, and spatial smoothing. An example of how our synthesis/analysis model can be used in validating an fMRI experiment is demonstrated using simulations and "hybrid" fMRI data.

AB - Methods for testing and validating independent component analysis (ICA) results in fMRI are growing in importance as the popularity of this model for studying brain function increases. We introduce and apply a synthesis/analysis model for analyzing functional Magnetic Resonance Imaging (fMRI) data using ICA. Classes of signal types relevant to fMRI are described and a statistical approach for validation of simulation results is developed. Additionally, we propose an empirical version of our validation approach to test the performance of various ICA approaches in "hybrid" fMRI data, a mixture of real fMRI data and known (validatable) sources. The synthesis portion of the model assumes statistically independent spatial sources in the brain. We also assume that the fMRI scanner acquires overdetermined data such that there are more time points than hemodynamic brain sources. We propose several signal classes relevant to fMRI and discuss the properties of each. The analysis portion of the model includes several candidates for spatial smoothing, ICA algorithm, and data reduction. We use the Kullback-Leibler divergence between the estimated source distributions and the "true" distributions as a measure of the optimality of the final ICA decomposition. Using this model, we generate fMRI-like data and optimize the analysis stage as a function of ICA algorithm, data reduction scheme, and spatial smoothing. An example of how our synthesis/analysis model can be used in validating an fMRI experiment is demonstrated using simulations and "hybrid" fMRI data.

KW - Brain

KW - fMRI

KW - Independent component analysis

KW - Model

UR - http://www.scopus.com/inward/record.url?scp=3543062215&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=3543062215&partnerID=8YFLogxK

U2 - 10.1023/B:VLSI.0000027491.81326.7a

DO - 10.1023/B:VLSI.0000027491.81326.7a

M3 - Article

AN - SCOPUS:3543062215

VL - 37

SP - 281

EP - 291

JO - Journal of Signal Processing Systems

JF - Journal of Signal Processing Systems

SN - 1939-8018

IS - 2-3

ER -