Abstract
Multiple-segmentation-and-fusion method has been widely used for brain extraction, tissue segmentation, and region of interest (ROI) localization. However, such studies are hindered in practice by their computational complexity, mainly coming from the steps of template selection and template-to-subject nonlinear registration. In this study, we address these two issues and propose a novel learning-based meta-algorithm for MRI brain extraction. Specifically, we first use exemplars to represent the entire template library, and assign the most similar exemplar to the test subject. Second, a meta-algorithm combining two existing brain extraction algorithms (BET and BSE) is proposed to conduct multiple extractions directly on test subject. Effective parameter settings for the meta-algorithm are learned from the training data and propagated to subject through exemplars. We further develop a level-set based fusion method to combine multiple candidate extractions together with a closed smooth surface, for obtaining the final result. Experimental results show that, with only a small portion of subjects for training, the proposed method is able to produce more accurate and robust brain extraction results, at Jaccard Index of 0.956±0.010 on total 340 subjects under 6-fold cross validation, compared to those by the BET and BSE even using their best parameter combinations.
| Original language | English |
|---|---|
| Title of host publication | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
| Pages | 313-321 |
| Number of pages | 9 |
| Volume | 6893 LNCS |
| Edition | PART 3 |
| DOIs | |
| Publication status | Published - 2011 Oct 11 |
| Externally published | Yes |
| Event | 14th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2011 - Toronto, ON, Canada Duration: 2011 Sep 18 → 2011 Sep 22 |
Publication series
| Name | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
|---|---|
| Number | PART 3 |
| Volume | 6893 LNCS |
| ISSN (Print) | 03029743 |
| ISSN (Electronic) | 16113349 |
Other
| Other | 14th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2011 |
|---|---|
| Country | Canada |
| City | Toronto, ON |
| Period | 11/9/18 → 11/9/22 |
Fingerprint
ASJC Scopus subject areas
- Computer Science(all)
- Theoretical Computer Science
Cite this
Learning-based meta-algorithm for MRI brain extraction. / Shi, Feng; Wang, Li; Gilmore, John H.; Lin, Weili; Shen, Dinggang.
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 6893 LNCS PART 3. ed. 2011. p. 313-321 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 6893 LNCS, No. PART 3).Research output: Chapter in Book/Report/Conference proceeding › Conference contribution
}
TY - GEN
T1 - Learning-based meta-algorithm for MRI brain extraction
AU - Shi, Feng
AU - Wang, Li
AU - Gilmore, John H.
AU - Lin, Weili
AU - Shen, Dinggang
PY - 2011/10/11
Y1 - 2011/10/11
N2 - Multiple-segmentation-and-fusion method has been widely used for brain extraction, tissue segmentation, and region of interest (ROI) localization. However, such studies are hindered in practice by their computational complexity, mainly coming from the steps of template selection and template-to-subject nonlinear registration. In this study, we address these two issues and propose a novel learning-based meta-algorithm for MRI brain extraction. Specifically, we first use exemplars to represent the entire template library, and assign the most similar exemplar to the test subject. Second, a meta-algorithm combining two existing brain extraction algorithms (BET and BSE) is proposed to conduct multiple extractions directly on test subject. Effective parameter settings for the meta-algorithm are learned from the training data and propagated to subject through exemplars. We further develop a level-set based fusion method to combine multiple candidate extractions together with a closed smooth surface, for obtaining the final result. Experimental results show that, with only a small portion of subjects for training, the proposed method is able to produce more accurate and robust brain extraction results, at Jaccard Index of 0.956±0.010 on total 340 subjects under 6-fold cross validation, compared to those by the BET and BSE even using their best parameter combinations.
AB - Multiple-segmentation-and-fusion method has been widely used for brain extraction, tissue segmentation, and region of interest (ROI) localization. However, such studies are hindered in practice by their computational complexity, mainly coming from the steps of template selection and template-to-subject nonlinear registration. In this study, we address these two issues and propose a novel learning-based meta-algorithm for MRI brain extraction. Specifically, we first use exemplars to represent the entire template library, and assign the most similar exemplar to the test subject. Second, a meta-algorithm combining two existing brain extraction algorithms (BET and BSE) is proposed to conduct multiple extractions directly on test subject. Effective parameter settings for the meta-algorithm are learned from the training data and propagated to subject through exemplars. We further develop a level-set based fusion method to combine multiple candidate extractions together with a closed smooth surface, for obtaining the final result. Experimental results show that, with only a small portion of subjects for training, the proposed method is able to produce more accurate and robust brain extraction results, at Jaccard Index of 0.956±0.010 on total 340 subjects under 6-fold cross validation, compared to those by the BET and BSE even using their best parameter combinations.
UR - http://www.scopus.com/inward/record.url?scp=80053493942&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=80053493942&partnerID=8YFLogxK
U2 - 10.1007/978-3-642-23626-6_39
DO - 10.1007/978-3-642-23626-6_39
M3 - Conference contribution
C2 - 22003714
AN - SCOPUS:80053493942
SN - 9783642236259
VL - 6893 LNCS
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 313
EP - 321
BT - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
ER -