A point says a lot: An interactive segmentation method for MR prostate via one-point labeling

Jinquan Sun, Yinghuan Shi, Yang Gao, Dinggang Shen

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

2 Citations (Scopus)

Abstract

In this paper, we investigate if the MR prostate segmentation performance could be improved, by only providing one-point labeling information in the prostate region. To achieve this goal, by asking the physician to first click one point inside the prostate region, we present a novel segmentation method by simultaneously integrating the boundary detection results and the patch-based prediction. Particularly, since the clicked point belongs to the prostate, we first generate the location-prior maps, with two basic assumptions: (1) a point closer to the clicked point should be with higher probability to be the prostate voxel, (2) a point separated by more boundaries to the clicked point, will have lower chance to be the prostate voxel. We perform the Canny edge detector and obtain two location-prior maps from horizontal and vertical directions, respectively. Then, the obtained location-prior maps along with the original MR images are fed into a multi-channel fully convolutional network to conduct the patch-based prediction. With the obtained prostate-likelihood map, we employ a level-set method to achieve the final segmentation. We evaluate the performance of our method on 22 MR images collected from 22 different patients, with the manual delineation provided as the ground truth for evaluation. The experimental results not only show the promising performance of our method but also demonstrate the one-point labeling could largely enhance the results when a pure patch-based prediction fails.

Original languageEnglish
Title of host publicationMachine Learning in Medical Imaging - 8th International Workshop, MLMI 2017, Held in Conjunction with MICCAI 2017, Proceedings
PublisherSpringer Verlag
Pages220-228
Number of pages9
Volume10541 LNCS
ISBN (Print)9783319673882
DOIs
Publication statusPublished - 2017 Jan 1
Externally publishedYes
Event8th International Workshop on Machine Learning in Medical Imaging, MLMI 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 102017 Sep 10

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10541 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other8th International Workshop on Machine Learning in Medical Imaging, MLMI 2017 held in conjunction with the 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017
CountryCanada
CityQuebec City
Period17/9/1017/9/10

Fingerprint

Labeling
Segmentation
Patch
Voxel
Prediction
Detectors
Boundary Detection
Level Set Method
Likelihood
Horizontal
Vertical
Detector
Evaluate
Evaluation
Experimental Results
Demonstrate

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Sun, J., Shi, Y., Gao, Y., & Shen, D. (2017). A point says a lot: An interactive segmentation method for MR prostate via one-point labeling. In Machine Learning in Medical Imaging - 8th International Workshop, MLMI 2017, Held in Conjunction with MICCAI 2017, Proceedings (Vol. 10541 LNCS, pp. 220-228). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10541 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-319-67389-9_26

A point says a lot : An interactive segmentation method for MR prostate via one-point labeling. / Sun, Jinquan; Shi, Yinghuan; Gao, Yang; Shen, Dinggang.

Machine Learning in Medical Imaging - 8th International Workshop, MLMI 2017, Held in Conjunction with MICCAI 2017, Proceedings. Vol. 10541 LNCS Springer Verlag, 2017. p. 220-228 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10541 LNCS).

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

Sun, J, Shi, Y, Gao, Y & Shen, D 2017, A point says a lot: An interactive segmentation method for MR prostate via one-point labeling. in Machine Learning in Medical Imaging - 8th International Workshop, MLMI 2017, Held in Conjunction with MICCAI 2017, Proceedings. vol. 10541 LNCS, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10541 LNCS, Springer Verlag, pp. 220-228, 8th International Workshop on Machine Learning in Medical Imaging, MLMI 2017 held in conjunction with the 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017, Quebec City, Canada, 17/9/10. https://doi.org/10.1007/978-3-319-67389-9_26
Sun J, Shi Y, Gao Y, Shen D. A point says a lot: An interactive segmentation method for MR prostate via one-point labeling. In Machine Learning in Medical Imaging - 8th International Workshop, MLMI 2017, Held in Conjunction with MICCAI 2017, Proceedings. Vol. 10541 LNCS. Springer Verlag. 2017. p. 220-228. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-319-67389-9_26
Sun, Jinquan ; Shi, Yinghuan ; Gao, Yang ; Shen, Dinggang. / A point says a lot : An interactive segmentation method for MR prostate via one-point labeling. Machine Learning in Medical Imaging - 8th International Workshop, MLMI 2017, Held in Conjunction with MICCAI 2017, Proceedings. Vol. 10541 LNCS Springer Verlag, 2017. pp. 220-228 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{1c79d7ed548d4a90b31287e19153aa55,
title = "A point says a lot: An interactive segmentation method for MR prostate via one-point labeling",
abstract = "In this paper, we investigate if the MR prostate segmentation performance could be improved, by only providing one-point labeling information in the prostate region. To achieve this goal, by asking the physician to first click one point inside the prostate region, we present a novel segmentation method by simultaneously integrating the boundary detection results and the patch-based prediction. Particularly, since the clicked point belongs to the prostate, we first generate the location-prior maps, with two basic assumptions: (1) a point closer to the clicked point should be with higher probability to be the prostate voxel, (2) a point separated by more boundaries to the clicked point, will have lower chance to be the prostate voxel. We perform the Canny edge detector and obtain two location-prior maps from horizontal and vertical directions, respectively. Then, the obtained location-prior maps along with the original MR images are fed into a multi-channel fully convolutional network to conduct the patch-based prediction. With the obtained prostate-likelihood map, we employ a level-set method to achieve the final segmentation. We evaluate the performance of our method on 22 MR images collected from 22 different patients, with the manual delineation provided as the ground truth for evaluation. The experimental results not only show the promising performance of our method but also demonstrate the one-point labeling could largely enhance the results when a pure patch-based prediction fails.",
author = "Jinquan Sun and Yinghuan Shi and Yang Gao and Dinggang Shen",
year = "2017",
month = "1",
day = "1",
doi = "10.1007/978-3-319-67389-9_26",
language = "English",
isbn = "9783319673882",
volume = "10541 LNCS",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "220--228",
booktitle = "Machine Learning in Medical Imaging - 8th International Workshop, MLMI 2017, Held in Conjunction with MICCAI 2017, Proceedings",

}

TY - GEN

T1 - A point says a lot

T2 - An interactive segmentation method for MR prostate via one-point labeling

AU - Sun, Jinquan

AU - Shi, Yinghuan

AU - Gao, Yang

AU - Shen, Dinggang

PY - 2017/1/1

Y1 - 2017/1/1

N2 - In this paper, we investigate if the MR prostate segmentation performance could be improved, by only providing one-point labeling information in the prostate region. To achieve this goal, by asking the physician to first click one point inside the prostate region, we present a novel segmentation method by simultaneously integrating the boundary detection results and the patch-based prediction. Particularly, since the clicked point belongs to the prostate, we first generate the location-prior maps, with two basic assumptions: (1) a point closer to the clicked point should be with higher probability to be the prostate voxel, (2) a point separated by more boundaries to the clicked point, will have lower chance to be the prostate voxel. We perform the Canny edge detector and obtain two location-prior maps from horizontal and vertical directions, respectively. Then, the obtained location-prior maps along with the original MR images are fed into a multi-channel fully convolutional network to conduct the patch-based prediction. With the obtained prostate-likelihood map, we employ a level-set method to achieve the final segmentation. We evaluate the performance of our method on 22 MR images collected from 22 different patients, with the manual delineation provided as the ground truth for evaluation. The experimental results not only show the promising performance of our method but also demonstrate the one-point labeling could largely enhance the results when a pure patch-based prediction fails.

AB - In this paper, we investigate if the MR prostate segmentation performance could be improved, by only providing one-point labeling information in the prostate region. To achieve this goal, by asking the physician to first click one point inside the prostate region, we present a novel segmentation method by simultaneously integrating the boundary detection results and the patch-based prediction. Particularly, since the clicked point belongs to the prostate, we first generate the location-prior maps, with two basic assumptions: (1) a point closer to the clicked point should be with higher probability to be the prostate voxel, (2) a point separated by more boundaries to the clicked point, will have lower chance to be the prostate voxel. We perform the Canny edge detector and obtain two location-prior maps from horizontal and vertical directions, respectively. Then, the obtained location-prior maps along with the original MR images are fed into a multi-channel fully convolutional network to conduct the patch-based prediction. With the obtained prostate-likelihood map, we employ a level-set method to achieve the final segmentation. We evaluate the performance of our method on 22 MR images collected from 22 different patients, with the manual delineation provided as the ground truth for evaluation. The experimental results not only show the promising performance of our method but also demonstrate the one-point labeling could largely enhance the results when a pure patch-based prediction fails.

UR - http://www.scopus.com/inward/record.url?scp=85029716704&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85029716704&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-67389-9_26

DO - 10.1007/978-3-319-67389-9_26

M3 - Conference contribution

AN - SCOPUS:85029716704

SN - 9783319673882

VL - 10541 LNCS

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

SP - 220

EP - 228

BT - Machine Learning in Medical Imaging - 8th International Workshop, MLMI 2017, Held in Conjunction with MICCAI 2017, Proceedings

PB - Springer Verlag

ER -