Segmenting new image acquisitions without labels

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

Abstract

We are interested in solving the problem of segmentation when no gold-standard labels are available for new image acquisition protocols. We developed a dual generative adversarial network (GAN), called Synth-GAN, which incorporates a differential operator loss (to favor retaining edges), as well as cyclic loss (to guarantee reconstruction of inputs). We show how the developed approach facilitates the application of an automated deep learning approach trained on one type of image (T2-weighted fat-sat MR) to be successfully applied to images well outside the trained distribution (Tl-weighted MR). A total of 100 images of each sequence from different patients were used (80% for training), and performance of the method was assessed by comparing how the previously developed automated segmentation approach performed prior to and post application of Synth-GAN. The developed approach improved the DICE coefficient from 0.39 (applying the automated segmentation method to the original Tl images) to 0.74 (applying the segmentation method to the synthesized T2 images). This approach will be useful for generalizing automated approaches across modalities, and institutions, when differences in hardware and software significantly alter image representations.

Original languageEnglish (US)
Title of host publicationISBI 2019 - 2019 IEEE International Symposium on Biomedical Imaging
PublisherIEEE Computer Society
Pages330-333
Number of pages4
ISBN (Electronic)9781538636411
DOIs
StatePublished - Apr 2019
Event16th IEEE International Symposium on Biomedical Imaging, ISBI 2019 - Venice, Italy
Duration: Apr 8 2019Apr 11 2019

Publication series

NameProceedings - International Symposium on Biomedical Imaging
Volume2019-April
ISSN (Print)1945-7928
ISSN (Electronic)1945-8452

Conference

Conference16th IEEE International Symposium on Biomedical Imaging, ISBI 2019
CountryItaly
CityVenice
Period4/8/194/11/19

Fingerprint

Image acquisition
Labels
Oils and fats
Gold
Hardware
Software
Fats
Learning
Deep learning

Keywords

  • Deep learning
  • Generative adversarial networks
  • Loss functions
  • Polycystic kidney disease
  • Total kidney volume

ASJC Scopus subject areas

  • Biomedical Engineering
  • Radiology Nuclear Medicine and imaging

Cite this

Kline, T. L. (2019). Segmenting new image acquisitions without labels. In ISBI 2019 - 2019 IEEE International Symposium on Biomedical Imaging (pp. 330-333). [8759175] (Proceedings - International Symposium on Biomedical Imaging; Vol. 2019-April). IEEE Computer Society. https://doi.org/10.1109/ISBI.2019.8759175

Segmenting new image acquisitions without labels. / Kline, Timothy L.

ISBI 2019 - 2019 IEEE International Symposium on Biomedical Imaging. IEEE Computer Society, 2019. p. 330-333 8759175 (Proceedings - International Symposium on Biomedical Imaging; Vol. 2019-April).

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

Kline, TL 2019, Segmenting new image acquisitions without labels. in ISBI 2019 - 2019 IEEE International Symposium on Biomedical Imaging., 8759175, Proceedings - International Symposium on Biomedical Imaging, vol. 2019-April, IEEE Computer Society, pp. 330-333, 16th IEEE International Symposium on Biomedical Imaging, ISBI 2019, Venice, Italy, 4/8/19. https://doi.org/10.1109/ISBI.2019.8759175
Kline TL. Segmenting new image acquisitions without labels. In ISBI 2019 - 2019 IEEE International Symposium on Biomedical Imaging. IEEE Computer Society. 2019. p. 330-333. 8759175. (Proceedings - International Symposium on Biomedical Imaging). https://doi.org/10.1109/ISBI.2019.8759175
Kline, Timothy L. / Segmenting new image acquisitions without labels. ISBI 2019 - 2019 IEEE International Symposium on Biomedical Imaging. IEEE Computer Society, 2019. pp. 330-333 (Proceedings - International Symposium on Biomedical Imaging).
@inproceedings{5f558364e13e48f3b18a55a2b97ed4cd,
title = "Segmenting new image acquisitions without labels",
abstract = "We are interested in solving the problem of segmentation when no gold-standard labels are available for new image acquisition protocols. We developed a dual generative adversarial network (GAN), called Synth-GAN, which incorporates a differential operator loss (to favor retaining edges), as well as cyclic loss (to guarantee reconstruction of inputs). We show how the developed approach facilitates the application of an automated deep learning approach trained on one type of image (T2-weighted fat-sat MR) to be successfully applied to images well outside the trained distribution (Tl-weighted MR). A total of 100 images of each sequence from different patients were used (80{\%} for training), and performance of the method was assessed by comparing how the previously developed automated segmentation approach performed prior to and post application of Synth-GAN. The developed approach improved the DICE coefficient from 0.39 (applying the automated segmentation method to the original Tl images) to 0.74 (applying the segmentation method to the synthesized T2 images). This approach will be useful for generalizing automated approaches across modalities, and institutions, when differences in hardware and software significantly alter image representations.",
keywords = "Deep learning, Generative adversarial networks, Loss functions, Polycystic kidney disease, Total kidney volume",
author = "Kline, {Timothy L.}",
year = "2019",
month = "4",
doi = "10.1109/ISBI.2019.8759175",
language = "English (US)",
series = "Proceedings - International Symposium on Biomedical Imaging",
publisher = "IEEE Computer Society",
pages = "330--333",
booktitle = "ISBI 2019 - 2019 IEEE International Symposium on Biomedical Imaging",

}

TY - GEN

T1 - Segmenting new image acquisitions without labels

AU - Kline, Timothy L.

PY - 2019/4

Y1 - 2019/4

N2 - We are interested in solving the problem of segmentation when no gold-standard labels are available for new image acquisition protocols. We developed a dual generative adversarial network (GAN), called Synth-GAN, which incorporates a differential operator loss (to favor retaining edges), as well as cyclic loss (to guarantee reconstruction of inputs). We show how the developed approach facilitates the application of an automated deep learning approach trained on one type of image (T2-weighted fat-sat MR) to be successfully applied to images well outside the trained distribution (Tl-weighted MR). A total of 100 images of each sequence from different patients were used (80% for training), and performance of the method was assessed by comparing how the previously developed automated segmentation approach performed prior to and post application of Synth-GAN. The developed approach improved the DICE coefficient from 0.39 (applying the automated segmentation method to the original Tl images) to 0.74 (applying the segmentation method to the synthesized T2 images). This approach will be useful for generalizing automated approaches across modalities, and institutions, when differences in hardware and software significantly alter image representations.

AB - We are interested in solving the problem of segmentation when no gold-standard labels are available for new image acquisition protocols. We developed a dual generative adversarial network (GAN), called Synth-GAN, which incorporates a differential operator loss (to favor retaining edges), as well as cyclic loss (to guarantee reconstruction of inputs). We show how the developed approach facilitates the application of an automated deep learning approach trained on one type of image (T2-weighted fat-sat MR) to be successfully applied to images well outside the trained distribution (Tl-weighted MR). A total of 100 images of each sequence from different patients were used (80% for training), and performance of the method was assessed by comparing how the previously developed automated segmentation approach performed prior to and post application of Synth-GAN. The developed approach improved the DICE coefficient from 0.39 (applying the automated segmentation method to the original Tl images) to 0.74 (applying the segmentation method to the synthesized T2 images). This approach will be useful for generalizing automated approaches across modalities, and institutions, when differences in hardware and software significantly alter image representations.

KW - Deep learning

KW - Generative adversarial networks

KW - Loss functions

KW - Polycystic kidney disease

KW - Total kidney volume

UR - http://www.scopus.com/inward/record.url?scp=85073896673&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85073896673&partnerID=8YFLogxK

U2 - 10.1109/ISBI.2019.8759175

DO - 10.1109/ISBI.2019.8759175

M3 - Conference contribution

AN - SCOPUS:85073896673

T3 - Proceedings - International Symposium on Biomedical Imaging

SP - 330

EP - 333

BT - ISBI 2019 - 2019 IEEE International Symposium on Biomedical Imaging

PB - IEEE Computer Society

ER -