Dimensionality estimation for optimal detection of functional networks in BOLD fMRI data

Grigori Yourganov, Xu Chen, Ana S. Lukic, Cheryl L. Grady, Steven L. Small, Miles N. Wernick, Stephen C. Strother

Research output: Contribution to journalArticle

33 Citations (Scopus)

Abstract

Estimation of the intrinsic dimensionality of fMRI data is an important part of data analysis that helps to separate the signal of interest from noise. We have studied multiple methods of dimensionality estimation proposed in the literature and used these estimates to select a subset of principal components that was subsequently processed by linear discriminant analysis (LDA). Using simulated multivariate Gaussian data, we show that the dimensionality that optimizes signal detection (in terms of the receiver operating characteristic (ROC) metric) goes through a transition from many dimensions to a single dimension as a function of the signal-to-noise ratio. This transition happens when the loci of activation are organized into a spatial network and the variance of the networked, task-related signals is high enough for the signal to be easily detected in the data. We show that reproducibility of activation maps is a metric that captures this switch in intrinsic dimensionality. Except for reproducibility, all of the methods of dimensionality estimation we considered failed to capture this transition: optimization of Bayesian evidence, minimum description length, supervised and unsupervised LDA prediction, and Stein's unbiased risk estimator. This failure results in sub-optimal ROC performance of LDA in the presence of a spatially distributed network, and may have caused LDA to underperform in many of the reported comparisons in the literature. Using real fMRI data sets, including multi-subject group and within-subject longitudinal analysis we demonstrate the existence of these dimensionality transitions in real data.

Original languageEnglish (US)
Pages (from-to)531-543
Number of pages13
JournalNeuroImage
Volume56
Issue number2
DOIs
StatePublished - May 15 2011
Externally publishedYes

Fingerprint

Discriminant Analysis
Magnetic Resonance Imaging
ROC Curve
Signal-To-Noise Ratio
Noise

Keywords

  • Dimensionality estimation
  • FMRI
  • Linear discriminant Analysis (LDA)
  • Model order selection
  • Principal component analysis (PCA)
  • Signal detection

ASJC Scopus subject areas

  • Neurology
  • Cognitive Neuroscience

Cite this

Yourganov, G., Chen, X., Lukic, A. S., Grady, C. L., Small, S. L., Wernick, M. N., & Strother, S. C. (2011). Dimensionality estimation for optimal detection of functional networks in BOLD fMRI data. NeuroImage, 56(2), 531-543. https://doi.org/10.1016/j.neuroimage.2010.09.034

Dimensionality estimation for optimal detection of functional networks in BOLD fMRI data. / Yourganov, Grigori; Chen, Xu; Lukic, Ana S.; Grady, Cheryl L.; Small, Steven L.; Wernick, Miles N.; Strother, Stephen C.

In: NeuroImage, Vol. 56, No. 2, 15.05.2011, p. 531-543.

Research output: Contribution to journalArticle

Yourganov, G, Chen, X, Lukic, AS, Grady, CL, Small, SL, Wernick, MN & Strother, SC 2011, 'Dimensionality estimation for optimal detection of functional networks in BOLD fMRI data', NeuroImage, vol. 56, no. 2, pp. 531-543. https://doi.org/10.1016/j.neuroimage.2010.09.034
Yourganov, Grigori ; Chen, Xu ; Lukic, Ana S. ; Grady, Cheryl L. ; Small, Steven L. ; Wernick, Miles N. ; Strother, Stephen C. / Dimensionality estimation for optimal detection of functional networks in BOLD fMRI data. In: NeuroImage. 2011 ; Vol. 56, No. 2. pp. 531-543.
@article{2a86c0875a5e447fb7fd51a51803c7e6,
title = "Dimensionality estimation for optimal detection of functional networks in BOLD fMRI data",
abstract = "Estimation of the intrinsic dimensionality of fMRI data is an important part of data analysis that helps to separate the signal of interest from noise. We have studied multiple methods of dimensionality estimation proposed in the literature and used these estimates to select a subset of principal components that was subsequently processed by linear discriminant analysis (LDA). Using simulated multivariate Gaussian data, we show that the dimensionality that optimizes signal detection (in terms of the receiver operating characteristic (ROC) metric) goes through a transition from many dimensions to a single dimension as a function of the signal-to-noise ratio. This transition happens when the loci of activation are organized into a spatial network and the variance of the networked, task-related signals is high enough for the signal to be easily detected in the data. We show that reproducibility of activation maps is a metric that captures this switch in intrinsic dimensionality. Except for reproducibility, all of the methods of dimensionality estimation we considered failed to capture this transition: optimization of Bayesian evidence, minimum description length, supervised and unsupervised LDA prediction, and Stein's unbiased risk estimator. This failure results in sub-optimal ROC performance of LDA in the presence of a spatially distributed network, and may have caused LDA to underperform in many of the reported comparisons in the literature. Using real fMRI data sets, including multi-subject group and within-subject longitudinal analysis we demonstrate the existence of these dimensionality transitions in real data.",
keywords = "Dimensionality estimation, FMRI, Linear discriminant Analysis (LDA), Model order selection, Principal component analysis (PCA), Signal detection",
author = "Grigori Yourganov and Xu Chen and Lukic, {Ana S.} and Grady, {Cheryl L.} and Small, {Steven L.} and Wernick, {Miles N.} and Strother, {Stephen C.}",
year = "2011",
month = "5",
day = "15",
doi = "10.1016/j.neuroimage.2010.09.034",
language = "English (US)",
volume = "56",
pages = "531--543",
journal = "NeuroImage",
issn = "1053-8119",
publisher = "Academic Press Inc.",
number = "2",

}

TY - JOUR

T1 - Dimensionality estimation for optimal detection of functional networks in BOLD fMRI data

AU - Yourganov, Grigori

AU - Chen, Xu

AU - Lukic, Ana S.

AU - Grady, Cheryl L.

AU - Small, Steven L.

AU - Wernick, Miles N.

AU - Strother, Stephen C.

PY - 2011/5/15

Y1 - 2011/5/15

N2 - Estimation of the intrinsic dimensionality of fMRI data is an important part of data analysis that helps to separate the signal of interest from noise. We have studied multiple methods of dimensionality estimation proposed in the literature and used these estimates to select a subset of principal components that was subsequently processed by linear discriminant analysis (LDA). Using simulated multivariate Gaussian data, we show that the dimensionality that optimizes signal detection (in terms of the receiver operating characteristic (ROC) metric) goes through a transition from many dimensions to a single dimension as a function of the signal-to-noise ratio. This transition happens when the loci of activation are organized into a spatial network and the variance of the networked, task-related signals is high enough for the signal to be easily detected in the data. We show that reproducibility of activation maps is a metric that captures this switch in intrinsic dimensionality. Except for reproducibility, all of the methods of dimensionality estimation we considered failed to capture this transition: optimization of Bayesian evidence, minimum description length, supervised and unsupervised LDA prediction, and Stein's unbiased risk estimator. This failure results in sub-optimal ROC performance of LDA in the presence of a spatially distributed network, and may have caused LDA to underperform in many of the reported comparisons in the literature. Using real fMRI data sets, including multi-subject group and within-subject longitudinal analysis we demonstrate the existence of these dimensionality transitions in real data.

AB - Estimation of the intrinsic dimensionality of fMRI data is an important part of data analysis that helps to separate the signal of interest from noise. We have studied multiple methods of dimensionality estimation proposed in the literature and used these estimates to select a subset of principal components that was subsequently processed by linear discriminant analysis (LDA). Using simulated multivariate Gaussian data, we show that the dimensionality that optimizes signal detection (in terms of the receiver operating characteristic (ROC) metric) goes through a transition from many dimensions to a single dimension as a function of the signal-to-noise ratio. This transition happens when the loci of activation are organized into a spatial network and the variance of the networked, task-related signals is high enough for the signal to be easily detected in the data. We show that reproducibility of activation maps is a metric that captures this switch in intrinsic dimensionality. Except for reproducibility, all of the methods of dimensionality estimation we considered failed to capture this transition: optimization of Bayesian evidence, minimum description length, supervised and unsupervised LDA prediction, and Stein's unbiased risk estimator. This failure results in sub-optimal ROC performance of LDA in the presence of a spatially distributed network, and may have caused LDA to underperform in many of the reported comparisons in the literature. Using real fMRI data sets, including multi-subject group and within-subject longitudinal analysis we demonstrate the existence of these dimensionality transitions in real data.

KW - Dimensionality estimation

KW - FMRI

KW - Linear discriminant Analysis (LDA)

KW - Model order selection

KW - Principal component analysis (PCA)

KW - Signal detection

UR - http://www.scopus.com/inward/record.url?scp=79955045640&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=79955045640&partnerID=8YFLogxK

U2 - 10.1016/j.neuroimage.2010.09.034

DO - 10.1016/j.neuroimage.2010.09.034

M3 - Article

C2 - 20858546

AN - SCOPUS:79955045640

VL - 56

SP - 531

EP - 543

JO - NeuroImage

JF - NeuroImage

SN - 1053-8119

IS - 2

ER -