Dual-layer ℓ1-graph embedding for semi-supervised image labeling

Qian Wang, Guorong Wu, Dinggang Shen

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

1 Citation (Scopus)

Abstract

In non-local patch-based (NLPB) labeling, a target voxel can fuse its label from the manual labels of the atlas voxels in accordance to the patch-based voxel similarities. Although state-of-the-art NLPB method mainly focuses on labeling a single target image by many atlases, we propose a novel semi-supervised strategy to address the realistic case of only a few atlases yet many unlabeled targets. Specifically, we create an ℓ1-graph of voxels, such that each target voxel can fuse its label from not only atlas voxels but also other target voxels. Meanwhile, each atlas voxel can utilize the feedbacks from the graph to check whether its expert labeling needs to be corrected. The ℓ1-graph is built by applying (duallayer) sparsity learning to all target and atlas voxels represented by their surrounding patches. By embedding the voxel labels to the graph, the target voxels can jointly compute their labels. In the experiment, our method with the capabilities of (1) joint labeling and (2) atlas label correction has enhanced the accuracy of NLPB labeling significantly.

Original languageEnglish
Title of host publicationLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
PublisherSpringer Verlag
Pages46-53
Number of pages8
Volume9467
ISBN (Print)9783319281933
DOIs
Publication statusPublished - 2015
Externally publishedYes
Event1st International Workshop on Patch-Based Techniques in Medical Imaging, Patch-MI 2015 - Munich, Germany
Duration: 2015 Oct 92015 Oct 9

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume9467
ISSN (Print)03029743
ISSN (Electronic)16113349

Other

Other1st International Workshop on Patch-Based Techniques in Medical Imaging, Patch-MI 2015
CountryGermany
CityMunich
Period15/10/915/10/9

Fingerprint

Graph Embedding
Voxel
Labeling
Labels
Atlas
Patch
Target
Electric fuses
Graph in graph theory
Feedback
Sparsity

ASJC Scopus subject areas

  • Computer Science(all)
  • Theoretical Computer Science

Cite this

Wang, Q., Wu, G., & Shen, D. (2015). Dual-layer ℓ1-graph embedding for semi-supervised image labeling. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 9467, pp. 46-53). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 9467). Springer Verlag. https://doi.org/10.1007/978-3-319-28194-0_6

Dual-layer ℓ1-graph embedding for semi-supervised image labeling. / Wang, Qian; Wu, Guorong; Shen, Dinggang.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 9467 Springer Verlag, 2015. p. 46-53 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 9467).

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

Wang, Q, Wu, G & Shen, D 2015, Dual-layer ℓ1-graph embedding for semi-supervised image labeling. in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). vol. 9467, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 9467, Springer Verlag, pp. 46-53, 1st International Workshop on Patch-Based Techniques in Medical Imaging, Patch-MI 2015, Munich, Germany, 15/10/9. https://doi.org/10.1007/978-3-319-28194-0_6
Wang Q, Wu G, Shen D. Dual-layer ℓ1-graph embedding for semi-supervised image labeling. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 9467. Springer Verlag. 2015. p. 46-53. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-319-28194-0_6
Wang, Qian ; Wu, Guorong ; Shen, Dinggang. / Dual-layer ℓ1-graph embedding for semi-supervised image labeling. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 9467 Springer Verlag, 2015. pp. 46-53 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{b833948b96d54a378d6d6fad0692a020,
title = "Dual-layer ℓ1-graph embedding for semi-supervised image labeling",
abstract = "In non-local patch-based (NLPB) labeling, a target voxel can fuse its label from the manual labels of the atlas voxels in accordance to the patch-based voxel similarities. Although state-of-the-art NLPB method mainly focuses on labeling a single target image by many atlases, we propose a novel semi-supervised strategy to address the realistic case of only a few atlases yet many unlabeled targets. Specifically, we create an ℓ1-graph of voxels, such that each target voxel can fuse its label from not only atlas voxels but also other target voxels. Meanwhile, each atlas voxel can utilize the feedbacks from the graph to check whether its expert labeling needs to be corrected. The ℓ1-graph is built by applying (duallayer) sparsity learning to all target and atlas voxels represented by their surrounding patches. By embedding the voxel labels to the graph, the target voxels can jointly compute their labels. In the experiment, our method with the capabilities of (1) joint labeling and (2) atlas label correction has enhanced the accuracy of NLPB labeling significantly.",
author = "Qian Wang and Guorong Wu and Dinggang Shen",
year = "2015",
doi = "10.1007/978-3-319-28194-0_6",
language = "English",
isbn = "9783319281933",
volume = "9467",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "46--53",
booktitle = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

}

TY - GEN

T1 - Dual-layer ℓ1-graph embedding for semi-supervised image labeling

AU - Wang, Qian

AU - Wu, Guorong

AU - Shen, Dinggang

PY - 2015

Y1 - 2015

N2 - In non-local patch-based (NLPB) labeling, a target voxel can fuse its label from the manual labels of the atlas voxels in accordance to the patch-based voxel similarities. Although state-of-the-art NLPB method mainly focuses on labeling a single target image by many atlases, we propose a novel semi-supervised strategy to address the realistic case of only a few atlases yet many unlabeled targets. Specifically, we create an ℓ1-graph of voxels, such that each target voxel can fuse its label from not only atlas voxels but also other target voxels. Meanwhile, each atlas voxel can utilize the feedbacks from the graph to check whether its expert labeling needs to be corrected. The ℓ1-graph is built by applying (duallayer) sparsity learning to all target and atlas voxels represented by their surrounding patches. By embedding the voxel labels to the graph, the target voxels can jointly compute their labels. In the experiment, our method with the capabilities of (1) joint labeling and (2) atlas label correction has enhanced the accuracy of NLPB labeling significantly.

AB - In non-local patch-based (NLPB) labeling, a target voxel can fuse its label from the manual labels of the atlas voxels in accordance to the patch-based voxel similarities. Although state-of-the-art NLPB method mainly focuses on labeling a single target image by many atlases, we propose a novel semi-supervised strategy to address the realistic case of only a few atlases yet many unlabeled targets. Specifically, we create an ℓ1-graph of voxels, such that each target voxel can fuse its label from not only atlas voxels but also other target voxels. Meanwhile, each atlas voxel can utilize the feedbacks from the graph to check whether its expert labeling needs to be corrected. The ℓ1-graph is built by applying (duallayer) sparsity learning to all target and atlas voxels represented by their surrounding patches. By embedding the voxel labels to the graph, the target voxels can jointly compute their labels. In the experiment, our method with the capabilities of (1) joint labeling and (2) atlas label correction has enhanced the accuracy of NLPB labeling significantly.

UR - http://www.scopus.com/inward/record.url?scp=84955303139&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84955303139&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-28194-0_6

DO - 10.1007/978-3-319-28194-0_6

M3 - Conference contribution

AN - SCOPUS:84955303139

SN - 9783319281933

VL - 9467

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

SP - 46

EP - 53

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

PB - Springer Verlag

ER -