Restricted Boltzmann machines for neuroimaging: An application in identifying intrinsic networks

R. Devon Hjelm, Vince Daniel Calhoun, Ruslan Salakhutdinov, Elena A. Allen, Tulay Adali, Sergey M. Plis

Research output: Contribution to journalArticle

Abstract

Matrix factorization models are the current dominant approach for resolving meaningful data-driven features in neuroimaging data. Among them, independent component analysis (ICA) is arguably the most widely used for identifying functional networks, and its success has led to a number of versatile extensions to group and multimodal data. However there are indications that ICA may have reached a limit in flexibility and representational capacity, as the majority of such extensions are case-driven, custom-made solutions that are still contained within the class of mixture models. In this work, we seek out a principled and naturally extensible approach and consider a probabilistic model known as a restricted Boltzmann machine (RBM). An RBM separates linear factors from functional brain imaging data by fitting a probability distribution model to the data. Importantly, the solution can be used as a building block for more complex (deep) models, making it naturally suitable for hierarchical and multimodal extensions that are not easily captured when using linear factorizations alone. We investigate the capability of RBMs to identify intrinsic networks and compare its performance to that of well-known linear mixture models, in particular ICA. Using synthetic and real task fMRI data, we show that RBMs can be used to identify networks and their temporal activations with accuracy that is equal or greater than that of factorization models. The demonstrated effectiveness of RBMs supports its use as a building block for deeper models, a significant prospect for future neuroimaging research.

Original languageEnglish (US)
Pages (from-to)245-260
Number of pages16
JournalNeuroImage
Volume96
DOIs
StatePublished - Aug 1 2014
Externally publishedYes

Fingerprint

Neuroimaging
Functional Neuroimaging
Statistical Models
Linear Models
Magnetic Resonance Imaging
Research

Keywords

  • DBN
  • FMRI
  • ICA
  • Intrinsic networks
  • RBM

ASJC Scopus subject areas

  • Cognitive Neuroscience
  • Neurology
  • Medicine(all)

Cite this

Restricted Boltzmann machines for neuroimaging : An application in identifying intrinsic networks. / Hjelm, R. Devon; Calhoun, Vince Daniel; Salakhutdinov, Ruslan; Allen, Elena A.; Adali, Tulay; Plis, Sergey M.

In: NeuroImage, Vol. 96, 01.08.2014, p. 245-260.

Research output: Contribution to journalArticle

Hjelm, R. Devon ; Calhoun, Vince Daniel ; Salakhutdinov, Ruslan ; Allen, Elena A. ; Adali, Tulay ; Plis, Sergey M. / Restricted Boltzmann machines for neuroimaging : An application in identifying intrinsic networks. In: NeuroImage. 2014 ; Vol. 96. pp. 245-260.
@article{fc567a42a065450ab2f00800d6ff0999,
title = "Restricted Boltzmann machines for neuroimaging: An application in identifying intrinsic networks",
abstract = "Matrix factorization models are the current dominant approach for resolving meaningful data-driven features in neuroimaging data. Among them, independent component analysis (ICA) is arguably the most widely used for identifying functional networks, and its success has led to a number of versatile extensions to group and multimodal data. However there are indications that ICA may have reached a limit in flexibility and representational capacity, as the majority of such extensions are case-driven, custom-made solutions that are still contained within the class of mixture models. In this work, we seek out a principled and naturally extensible approach and consider a probabilistic model known as a restricted Boltzmann machine (RBM). An RBM separates linear factors from functional brain imaging data by fitting a probability distribution model to the data. Importantly, the solution can be used as a building block for more complex (deep) models, making it naturally suitable for hierarchical and multimodal extensions that are not easily captured when using linear factorizations alone. We investigate the capability of RBMs to identify intrinsic networks and compare its performance to that of well-known linear mixture models, in particular ICA. Using synthetic and real task fMRI data, we show that RBMs can be used to identify networks and their temporal activations with accuracy that is equal or greater than that of factorization models. The demonstrated effectiveness of RBMs supports its use as a building block for deeper models, a significant prospect for future neuroimaging research.",
keywords = "DBN, FMRI, ICA, Intrinsic networks, RBM",
author = "Hjelm, {R. Devon} and Calhoun, {Vince Daniel} and Ruslan Salakhutdinov and Allen, {Elena A.} and Tulay Adali and Plis, {Sergey M.}",
year = "2014",
month = "8",
day = "1",
doi = "10.1016/j.neuroimage.2014.03.048",
language = "English (US)",
volume = "96",
pages = "245--260",
journal = "NeuroImage",
issn = "1053-8119",
publisher = "Academic Press Inc.",

}

TY - JOUR

T1 - Restricted Boltzmann machines for neuroimaging

T2 - An application in identifying intrinsic networks

AU - Hjelm, R. Devon

AU - Calhoun, Vince Daniel

AU - Salakhutdinov, Ruslan

AU - Allen, Elena A.

AU - Adali, Tulay

AU - Plis, Sergey M.

PY - 2014/8/1

Y1 - 2014/8/1

N2 - Matrix factorization models are the current dominant approach for resolving meaningful data-driven features in neuroimaging data. Among them, independent component analysis (ICA) is arguably the most widely used for identifying functional networks, and its success has led to a number of versatile extensions to group and multimodal data. However there are indications that ICA may have reached a limit in flexibility and representational capacity, as the majority of such extensions are case-driven, custom-made solutions that are still contained within the class of mixture models. In this work, we seek out a principled and naturally extensible approach and consider a probabilistic model known as a restricted Boltzmann machine (RBM). An RBM separates linear factors from functional brain imaging data by fitting a probability distribution model to the data. Importantly, the solution can be used as a building block for more complex (deep) models, making it naturally suitable for hierarchical and multimodal extensions that are not easily captured when using linear factorizations alone. We investigate the capability of RBMs to identify intrinsic networks and compare its performance to that of well-known linear mixture models, in particular ICA. Using synthetic and real task fMRI data, we show that RBMs can be used to identify networks and their temporal activations with accuracy that is equal or greater than that of factorization models. The demonstrated effectiveness of RBMs supports its use as a building block for deeper models, a significant prospect for future neuroimaging research.

AB - Matrix factorization models are the current dominant approach for resolving meaningful data-driven features in neuroimaging data. Among them, independent component analysis (ICA) is arguably the most widely used for identifying functional networks, and its success has led to a number of versatile extensions to group and multimodal data. However there are indications that ICA may have reached a limit in flexibility and representational capacity, as the majority of such extensions are case-driven, custom-made solutions that are still contained within the class of mixture models. In this work, we seek out a principled and naturally extensible approach and consider a probabilistic model known as a restricted Boltzmann machine (RBM). An RBM separates linear factors from functional brain imaging data by fitting a probability distribution model to the data. Importantly, the solution can be used as a building block for more complex (deep) models, making it naturally suitable for hierarchical and multimodal extensions that are not easily captured when using linear factorizations alone. We investigate the capability of RBMs to identify intrinsic networks and compare its performance to that of well-known linear mixture models, in particular ICA. Using synthetic and real task fMRI data, we show that RBMs can be used to identify networks and their temporal activations with accuracy that is equal or greater than that of factorization models. The demonstrated effectiveness of RBMs supports its use as a building block for deeper models, a significant prospect for future neuroimaging research.

KW - DBN

KW - FMRI

KW - ICA

KW - Intrinsic networks

KW - RBM

UR - http://www.scopus.com/inward/record.url?scp=84899931628&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84899931628&partnerID=8YFLogxK

U2 - 10.1016/j.neuroimage.2014.03.048

DO - 10.1016/j.neuroimage.2014.03.048

M3 - Article

C2 - 24680869

AN - SCOPUS:84899931628

VL - 96

SP - 245

EP - 260

JO - NeuroImage

JF - NeuroImage

SN - 1053-8119

ER -