Control-group feature normalization for multivariate pattern analysis of structural MRI data using the support vector machine

Kristin A. Linn, Bilwaj Gaonkar, Theodore D. Satterthwaite, Jimit Doshi, Christos Davatzikos, Russell T. Shinohara

Research output: Contribution to journalArticle

Abstract

Normalization of feature vector values is a common practice in machine learning. Generally, each feature value is standardized to the unit hypercube or by normalizing to zero mean and unit variance. Classification decisions based on support vector machines (SVMs) or by other methods are sensitive to the specific normalization used on the features. In the context of multivariate pattern analysis using neuroimaging data, standardization effectively up- and down-weights features based on their individual variability. Since the standard approach uses the entire data set to guide the normalization, it utilizes the total variability of these features. This total variation is inevitably dependent on the amount of marginal separation between groups. Thus, such a normalization may attenuate the separability of the data in high dimensional space. In this work we propose an alternate approach that uses an estimate of the control-group standard deviation to normalize features before training. We study our proposed approach in the context of group classification using structural MRI data. We show that control-based normalization leads to better reproducibility of estimated multivariate disease patterns and improves the classifier performance in many cases.

Original languageEnglish (US)
Pages (from-to)157-166
Number of pages10
JournalNeuroImage
Volume132
DOIs
StatePublished - May 15 2016
Externally publishedYes

Fingerprint

Multivariate Analysis
Weights and Measures
Control Groups
Neuroimaging
Support Vector Machine
Datasets
Machine Learning

Keywords

  • Feature normalization
  • Multivariate pattern analysis
  • Structural MRI
  • Support vector machine

ASJC Scopus subject areas

  • Cognitive Neuroscience
  • Neurology

Cite this

Control-group feature normalization for multivariate pattern analysis of structural MRI data using the support vector machine. / Linn, Kristin A.; Gaonkar, Bilwaj; Satterthwaite, Theodore D.; Doshi, Jimit; Davatzikos, Christos; Shinohara, Russell T.

In: NeuroImage, Vol. 132, 15.05.2016, p. 157-166.

Research output: Contribution to journalArticle

Linn, Kristin A. ; Gaonkar, Bilwaj ; Satterthwaite, Theodore D. ; Doshi, Jimit ; Davatzikos, Christos ; Shinohara, Russell T. / Control-group feature normalization for multivariate pattern analysis of structural MRI data using the support vector machine. In: NeuroImage. 2016 ; Vol. 132. pp. 157-166.
@article{c1219defc78f4179a6f6b8a0d6978d47,
title = "Control-group feature normalization for multivariate pattern analysis of structural MRI data using the support vector machine",
abstract = "Normalization of feature vector values is a common practice in machine learning. Generally, each feature value is standardized to the unit hypercube or by normalizing to zero mean and unit variance. Classification decisions based on support vector machines (SVMs) or by other methods are sensitive to the specific normalization used on the features. In the context of multivariate pattern analysis using neuroimaging data, standardization effectively up- and down-weights features based on their individual variability. Since the standard approach uses the entire data set to guide the normalization, it utilizes the total variability of these features. This total variation is inevitably dependent on the amount of marginal separation between groups. Thus, such a normalization may attenuate the separability of the data in high dimensional space. In this work we propose an alternate approach that uses an estimate of the control-group standard deviation to normalize features before training. We study our proposed approach in the context of group classification using structural MRI data. We show that control-based normalization leads to better reproducibility of estimated multivariate disease patterns and improves the classifier performance in many cases.",
keywords = "Feature normalization, Multivariate pattern analysis, Structural MRI, Support vector machine",
author = "Linn, {Kristin A.} and Bilwaj Gaonkar and Satterthwaite, {Theodore D.} and Jimit Doshi and Christos Davatzikos and Shinohara, {Russell T.}",
year = "2016",
month = "5",
day = "15",
doi = "10.1016/j.neuroimage.2016.02.044",
language = "English (US)",
volume = "132",
pages = "157--166",
journal = "NeuroImage",
issn = "1053-8119",
publisher = "Academic Press Inc.",

}

TY - JOUR

T1 - Control-group feature normalization for multivariate pattern analysis of structural MRI data using the support vector machine

AU - Linn, Kristin A.

AU - Gaonkar, Bilwaj

AU - Satterthwaite, Theodore D.

AU - Doshi, Jimit

AU - Davatzikos, Christos

AU - Shinohara, Russell T.

PY - 2016/5/15

Y1 - 2016/5/15

N2 - Normalization of feature vector values is a common practice in machine learning. Generally, each feature value is standardized to the unit hypercube or by normalizing to zero mean and unit variance. Classification decisions based on support vector machines (SVMs) or by other methods are sensitive to the specific normalization used on the features. In the context of multivariate pattern analysis using neuroimaging data, standardization effectively up- and down-weights features based on their individual variability. Since the standard approach uses the entire data set to guide the normalization, it utilizes the total variability of these features. This total variation is inevitably dependent on the amount of marginal separation between groups. Thus, such a normalization may attenuate the separability of the data in high dimensional space. In this work we propose an alternate approach that uses an estimate of the control-group standard deviation to normalize features before training. We study our proposed approach in the context of group classification using structural MRI data. We show that control-based normalization leads to better reproducibility of estimated multivariate disease patterns and improves the classifier performance in many cases.

AB - Normalization of feature vector values is a common practice in machine learning. Generally, each feature value is standardized to the unit hypercube or by normalizing to zero mean and unit variance. Classification decisions based on support vector machines (SVMs) or by other methods are sensitive to the specific normalization used on the features. In the context of multivariate pattern analysis using neuroimaging data, standardization effectively up- and down-weights features based on their individual variability. Since the standard approach uses the entire data set to guide the normalization, it utilizes the total variability of these features. This total variation is inevitably dependent on the amount of marginal separation between groups. Thus, such a normalization may attenuate the separability of the data in high dimensional space. In this work we propose an alternate approach that uses an estimate of the control-group standard deviation to normalize features before training. We study our proposed approach in the context of group classification using structural MRI data. We show that control-based normalization leads to better reproducibility of estimated multivariate disease patterns and improves the classifier performance in many cases.

KW - Feature normalization

KW - Multivariate pattern analysis

KW - Structural MRI

KW - Support vector machine

UR - http://www.scopus.com/inward/record.url?scp=84959316755&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84959316755&partnerID=8YFLogxK

U2 - 10.1016/j.neuroimage.2016.02.044

DO - 10.1016/j.neuroimage.2016.02.044

M3 - Article

VL - 132

SP - 157

EP - 166

JO - NeuroImage

JF - NeuroImage

SN - 1053-8119

ER -