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 language | English |
|---|---|
| Title of host publication | Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings |
| Editors | Leonardo Bonilha, Guorong Wu, Paul Laurienti, Brent C. Munsell |
| Publisher | Springer Verlag |
| Pages | 143-151 |
| Number of pages | 9 |
| ISBN (Print) | 9783319671581 |
| DOIs | |
| Publication status | Published - 2017 Jan 1 |
| Event | 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 Duration: 2017 Sep 14 → 2017 Sep 14 |
Publication series
| Name | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
|---|---|
| Volume | 10511 LNCS |
| ISSN (Print) | 0302-9743 |
| ISSN (Electronic) | 1611-3349 |
Other
| Other | 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 |
|---|---|
| Country | Canada |
| City | Quebec City |
| Period | 17/9/14 → 17/9/14 |
Fingerprint
ASJC Scopus subject areas
- Theoretical Computer Science
- Computer Science(all)
Cite this
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 proceeding › Conference contribution
}
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 -