Region-wise stochastic pattern modeling for autism spectrum disorder identification and temporal dynamics analysis

Eunji Jun, Heung-Il Suk

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

Abstract

Many studies in the literature have validated the use of resting-state fMRI (rs-fMRI) for brain disorder/disease identification. Unlike the existing methods that mostly first estimate functional connectivity and then extract features with a graph theory, in this paper, we propose a novel method that directly models the temporal stochastic patterns inherent in BOLD signals for each Region Of Interest (ROI) individually. Specifically, we model temporal BOLD signal fluctuation of an individual ROI by means of Hidden Markov Models (HMMs), and then compute a regional BOLD signal likelihood with the trained HMMs. By regarding the BOLD signal likelihood of ROIs over a whole brain as features, we build a classifier that can discriminate subjects with Autism Spectrum Disorder (ASD) from Normal healthy Controls (NC). In addition, we also devise a method to further investigate the characteristics of temporal dynamics in rs-fMRI estimated by HMMs. For group comparison, we use the metrics of state occupancy rate and lifetime of the optimal hidden states that best represent the temporal BOLD signals. In our experiments with ABIDE cohort, we validated the effectiveness of the proposed method by achieving the highest diagnostic accuracies among competing methods. We could also identify the group differences in temporal dynamics between ASD and NC in terms of state occupancy rate and lifetime of individual states.

Original languageEnglish
Title of host publicationConnectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings
EditorsLeonardo Bonilha, Guorong Wu, Paul Laurienti, Brent C. Munsell
PublisherSpringer Verlag
Pages143-151
Number of pages9
ISBN (Print)9783319671581
DOIs
Publication statusPublished - 2017 Jan 1
Event1st International Workshop on Connectomics in NeuroImaging, CNI 2017 held in conjunction with the 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017 - Quebec City, Canada
Duration: 2017 Sep 142017 Sep 14

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10511 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other1st International Workshop on Connectomics in NeuroImaging, CNI 2017 held in conjunction with the 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017
CountryCanada
CityQuebec City
Period17/9/1417/9/14

Fingerprint

Hidden Markov models
Dynamic Analysis
Dynamic analysis
Disorder
Brain
Markov Model
Modeling
Functional Magnetic Resonance Imaging
Region of Interest
Graph theory
Lifetime
Likelihood
Diagnostic Accuracy
Classifiers
High Accuracy
Connectivity
Classifier
Fluctuations
Metric
Experiments

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Jun, E., & Suk, H-I. (2017). Region-wise stochastic pattern modeling for autism spectrum disorder identification and temporal dynamics analysis. In L. Bonilha, G. Wu, P. Laurienti, & B. C. Munsell (Eds.), Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings (pp. 143-151). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10511 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-319-67159-8_17

Region-wise stochastic pattern modeling for autism spectrum disorder identification and temporal dynamics analysis. / Jun, Eunji; Suk, Heung-Il.

Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings. ed. / Leonardo Bonilha; Guorong Wu; Paul Laurienti; Brent C. Munsell. Springer Verlag, 2017. p. 143-151 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10511 LNCS).

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

Jun, E & Suk, H-I 2017, Region-wise stochastic pattern modeling for autism spectrum disorder identification and temporal dynamics analysis. in L Bonilha, G Wu, P Laurienti & BC Munsell (eds), Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10511 LNCS, Springer Verlag, pp. 143-151, 1st International Workshop on Connectomics in NeuroImaging, CNI 2017 held in conjunction with the 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017, Quebec City, Canada, 17/9/14. https://doi.org/10.1007/978-3-319-67159-8_17
Jun E, Suk H-I. Region-wise stochastic pattern modeling for autism spectrum disorder identification and temporal dynamics analysis. In Bonilha L, Wu G, Laurienti P, Munsell BC, editors, Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings. Springer Verlag. 2017. p. 143-151. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-319-67159-8_17
Jun, Eunji ; Suk, Heung-Il. / Region-wise stochastic pattern modeling for autism spectrum disorder identification and temporal dynamics analysis. Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings. editor / Leonardo Bonilha ; Guorong Wu ; Paul Laurienti ; Brent C. Munsell. Springer Verlag, 2017. pp. 143-151 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{54020de7edbf4c589eb9be339259a16c,
title = "Region-wise stochastic pattern modeling for autism spectrum disorder identification and temporal dynamics analysis",
abstract = "Many studies in the literature have validated the use of resting-state fMRI (rs-fMRI) for brain disorder/disease identification. Unlike the existing methods that mostly first estimate functional connectivity and then extract features with a graph theory, in this paper, we propose a novel method that directly models the temporal stochastic patterns inherent in BOLD signals for each Region Of Interest (ROI) individually. Specifically, we model temporal BOLD signal fluctuation of an individual ROI by means of Hidden Markov Models (HMMs), and then compute a regional BOLD signal likelihood with the trained HMMs. By regarding the BOLD signal likelihood of ROIs over a whole brain as features, we build a classifier that can discriminate subjects with Autism Spectrum Disorder (ASD) from Normal healthy Controls (NC). In addition, we also devise a method to further investigate the characteristics of temporal dynamics in rs-fMRI estimated by HMMs. For group comparison, we use the metrics of state occupancy rate and lifetime of the optimal hidden states that best represent the temporal BOLD signals. In our experiments with ABIDE cohort, we validated the effectiveness of the proposed method by achieving the highest diagnostic accuracies among competing methods. We could also identify the group differences in temporal dynamics between ASD and NC in terms of state occupancy rate and lifetime of individual states.",
author = "Eunji Jun and Heung-Il Suk",
year = "2017",
month = "1",
day = "1",
doi = "10.1007/978-3-319-67159-8_17",
language = "English",
isbn = "9783319671581",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "143--151",
editor = "Leonardo Bonilha and Guorong Wu and Paul Laurienti and Munsell, {Brent C.}",
booktitle = "Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings",

}

TY - GEN

T1 - Region-wise stochastic pattern modeling for autism spectrum disorder identification and temporal dynamics analysis

AU - Jun, Eunji

AU - Suk, Heung-Il

PY - 2017/1/1

Y1 - 2017/1/1

N2 - Many studies in the literature have validated the use of resting-state fMRI (rs-fMRI) for brain disorder/disease identification. Unlike the existing methods that mostly first estimate functional connectivity and then extract features with a graph theory, in this paper, we propose a novel method that directly models the temporal stochastic patterns inherent in BOLD signals for each Region Of Interest (ROI) individually. Specifically, we model temporal BOLD signal fluctuation of an individual ROI by means of Hidden Markov Models (HMMs), and then compute a regional BOLD signal likelihood with the trained HMMs. By regarding the BOLD signal likelihood of ROIs over a whole brain as features, we build a classifier that can discriminate subjects with Autism Spectrum Disorder (ASD) from Normal healthy Controls (NC). In addition, we also devise a method to further investigate the characteristics of temporal dynamics in rs-fMRI estimated by HMMs. For group comparison, we use the metrics of state occupancy rate and lifetime of the optimal hidden states that best represent the temporal BOLD signals. In our experiments with ABIDE cohort, we validated the effectiveness of the proposed method by achieving the highest diagnostic accuracies among competing methods. We could also identify the group differences in temporal dynamics between ASD and NC in terms of state occupancy rate and lifetime of individual states.

AB - Many studies in the literature have validated the use of resting-state fMRI (rs-fMRI) for brain disorder/disease identification. Unlike the existing methods that mostly first estimate functional connectivity and then extract features with a graph theory, in this paper, we propose a novel method that directly models the temporal stochastic patterns inherent in BOLD signals for each Region Of Interest (ROI) individually. Specifically, we model temporal BOLD signal fluctuation of an individual ROI by means of Hidden Markov Models (HMMs), and then compute a regional BOLD signal likelihood with the trained HMMs. By regarding the BOLD signal likelihood of ROIs over a whole brain as features, we build a classifier that can discriminate subjects with Autism Spectrum Disorder (ASD) from Normal healthy Controls (NC). In addition, we also devise a method to further investigate the characteristics of temporal dynamics in rs-fMRI estimated by HMMs. For group comparison, we use the metrics of state occupancy rate and lifetime of the optimal hidden states that best represent the temporal BOLD signals. In our experiments with ABIDE cohort, we validated the effectiveness of the proposed method by achieving the highest diagnostic accuracies among competing methods. We could also identify the group differences in temporal dynamics between ASD and NC in terms of state occupancy rate and lifetime of individual states.

UR - http://www.scopus.com/inward/record.url?scp=85029418048&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85029418048&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-67159-8_17

DO - 10.1007/978-3-319-67159-8_17

M3 - Conference contribution

AN - SCOPUS:85029418048

SN - 9783319671581

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 143

EP - 151

BT - Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings

A2 - Bonilha, Leonardo

A2 - Wu, Guorong

A2 - Laurienti, Paul

A2 - Munsell, Brent C.

PB - Springer Verlag

ER -