Deep learning from label proportions for emphysema quantification

Gerda Bortsova; Florian Dubost; Silas Ørting; Ioannis Katramados; Laurens Hogeweg; Laura Thomsen; Mathilde Wille; Marleen de Bruijne

doi:10.1007/978-3-030-00934-2_85

Deep learning from label proportions for emphysema quantification

Gerda Bortsova^*, Florian Dubost, Silas Ørting, Ioannis Katramados, Laurens Hogeweg, Laura Thomsen, Mathilde Wille, Marleen de Bruijne

^*Corresponding author for this work

Research output: Chapter/Conference proceeding › Conference proceeding › Academic › peer-review

23 Citations (Scopus)

Abstract

We propose an end-to-end deep learning method that learns to estimate emphysema extent from proportions of the diseased tissue. These proportions were visually estimated by experts using a standard grading system, in which grades correspond to intervals (label example: 1–5% of diseased tissue). The proposed architecture encodes the knowledge that the labels represent a volumetric proportion. A custom loss is designed to learn with intervals. Thus, during training, our network learns to segment the diseased tissue such that its proportions fit the ground truth intervals. Our architecture and loss combined improve the performance substantially (8% ICC) compared to a more conventional regression network. We outperform traditional lung densitometry and two recently published methods for emphysema quantification by a large margin (at least 7% AUC and 15% ICC), and achieve near-human-level performance. Moreover, our method generates emphysema segmentations that predict the spatial distribution of emphysema at human level.

Original language	English
Title of host publication	Medical Image Computing and Computer Assisted Intervention – MICCAI 2018 - 21st International Conference, 2018, Proceedings
Editors	Gabor Fichtinger, Christos Davatzikos, Carlos Alberola-López, Alejandro F. Frangi, Julia A. Schnabel
Pages	768-776
Number of pages	9
DOIs	https://doi.org/10.1007/978-3-030-00934-2_85
Publication status	Published - 2018
Event	21st International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2018 - Granada, Spain Duration: 16 Sept 2018 → 20 Sept 2018

Publication series

Series	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	11071 LNCS
ISSN	0302-9743

Conference

Conference	21st International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2018
Country/Territory	Spain
City	Granada
Period	16/09/18 → 20/09/18

Bibliographical note

Funding Information:
Acknowledgment. This research is financed by the Netherlands Organization for Scientific Research (NWO) and COSMONiO.

Publisher Copyright:
© Springer Nature Switzerland AG 2018.

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1007/978-3-030-00934-2_85

Cite this

Bortsova, G., Dubost, F., Ørting, S., Katramados, I., Hogeweg, L., Thomsen, L., Wille, M., & de Bruijne, M. (2018). Deep learning from label proportions for emphysema quantification. In G. Fichtinger, C. Davatzikos, C. Alberola-López, A. F. Frangi, & J. A. Schnabel (Eds.), Medical Image Computing and Computer Assisted Intervention – MICCAI 2018 - 21st International Conference, 2018, Proceedings (pp. 768-776) https://doi.org/10.1007/978-3-030-00934-2_85

Bortsova, Gerda ; Dubost, Florian ; Ørting, Silas et al. / Deep learning from label proportions for emphysema quantification. Medical Image Computing and Computer Assisted Intervention – MICCAI 2018 - 21st International Conference, 2018, Proceedings. editor / Gabor Fichtinger ; Christos Davatzikos ; Carlos Alberola-López ; Alejandro F. Frangi ; Julia A. Schnabel. 2018. pp. 768-776 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol. 11071 LNCS).

@inproceedings{0fe9678548314cad892fe95fc146dcf5,

title = "Deep learning from label proportions for emphysema quantification",

abstract = "We propose an end-to-end deep learning method that learns to estimate emphysema extent from proportions of the diseased tissue. These proportions were visually estimated by experts using a standard grading system, in which grades correspond to intervals (label example: 1–5% of diseased tissue). The proposed architecture encodes the knowledge that the labels represent a volumetric proportion. A custom loss is designed to learn with intervals. Thus, during training, our network learns to segment the diseased tissue such that its proportions fit the ground truth intervals. Our architecture and loss combined improve the performance substantially (8% ICC) compared to a more conventional regression network. We outperform traditional lung densitometry and two recently published methods for emphysema quantification by a large margin (at least 7% AUC and 15% ICC), and achieve near-human-level performance. Moreover, our method generates emphysema segmentations that predict the spatial distribution of emphysema at human level.",

author = "Gerda Bortsova and Florian Dubost and Silas {\O}rting and Ioannis Katramados and Laurens Hogeweg and Laura Thomsen and Mathilde Wille and {de Bruijne}, Marleen",

note = "Funding Information: Acknowledgment. This research is financed by the Netherlands Organization for Scientific Research (NWO) and COSMONiO. Publisher Copyright: {\textcopyright} Springer Nature Switzerland AG 2018.; 21st International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2018 ; Conference date: 16-09-2018 Through 20-09-2018",

year = "2018",

doi = "10.1007/978-3-030-00934-2_85",

language = "English",

isbn = "9783030009335",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer-Verlag",

pages = "768--776",

editor = "Gabor Fichtinger and Christos Davatzikos and Carlos Alberola-L{\'o}pez and Frangi, {Alejandro F.} and Schnabel, {Julia A.}",

booktitle = "Medical Image Computing and Computer Assisted Intervention – MICCAI 2018 - 21st International Conference, 2018, Proceedings",

}

Bortsova, G, Dubost, F, Ørting, S, Katramados, I, Hogeweg, L, Thomsen, L, Wille, M & de Bruijne, M 2018, Deep learning from label proportions for emphysema quantification. in G Fichtinger, C Davatzikos, C Alberola-López, AF Frangi & JA Schnabel (eds), Medical Image Computing and Computer Assisted Intervention – MICCAI 2018 - 21st International Conference, 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11071 LNCS, pp. 768-776, 21st International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2018, Granada, Spain, 16/09/18. https://doi.org/10.1007/978-3-030-00934-2_85

Deep learning from label proportions for emphysema quantification. / Bortsova, Gerda; Dubost, Florian; Ørting, Silas et al.
Medical Image Computing and Computer Assisted Intervention – MICCAI 2018 - 21st International Conference, 2018, Proceedings. ed. / Gabor Fichtinger; Christos Davatzikos; Carlos Alberola-López; Alejandro F. Frangi; Julia A. Schnabel. 2018. p. 768-776 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol. 11071 LNCS).

Research output: Chapter/Conference proceeding › Conference proceeding › Academic › peer-review

TY - GEN

T1 - Deep learning from label proportions for emphysema quantification

AU - Bortsova, Gerda

AU - Dubost, Florian

AU - Ørting, Silas

AU - Katramados, Ioannis

AU - Hogeweg, Laurens

AU - Thomsen, Laura

AU - Wille, Mathilde

AU - de Bruijne, Marleen

N1 - Funding Information: Acknowledgment. This research is financed by the Netherlands Organization for Scientific Research (NWO) and COSMONiO. Publisher Copyright: © Springer Nature Switzerland AG 2018.

PY - 2018

Y1 - 2018

N2 - We propose an end-to-end deep learning method that learns to estimate emphysema extent from proportions of the diseased tissue. These proportions were visually estimated by experts using a standard grading system, in which grades correspond to intervals (label example: 1–5% of diseased tissue). The proposed architecture encodes the knowledge that the labels represent a volumetric proportion. A custom loss is designed to learn with intervals. Thus, during training, our network learns to segment the diseased tissue such that its proportions fit the ground truth intervals. Our architecture and loss combined improve the performance substantially (8% ICC) compared to a more conventional regression network. We outperform traditional lung densitometry and two recently published methods for emphysema quantification by a large margin (at least 7% AUC and 15% ICC), and achieve near-human-level performance. Moreover, our method generates emphysema segmentations that predict the spatial distribution of emphysema at human level.

AB - We propose an end-to-end deep learning method that learns to estimate emphysema extent from proportions of the diseased tissue. These proportions were visually estimated by experts using a standard grading system, in which grades correspond to intervals (label example: 1–5% of diseased tissue). The proposed architecture encodes the knowledge that the labels represent a volumetric proportion. A custom loss is designed to learn with intervals. Thus, during training, our network learns to segment the diseased tissue such that its proportions fit the ground truth intervals. Our architecture and loss combined improve the performance substantially (8% ICC) compared to a more conventional regression network. We outperform traditional lung densitometry and two recently published methods for emphysema quantification by a large margin (at least 7% AUC and 15% ICC), and achieve near-human-level performance. Moreover, our method generates emphysema segmentations that predict the spatial distribution of emphysema at human level.

UR - http://www.scopus.com/inward/record.url?scp=85054065475&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-00934-2_85

DO - 10.1007/978-3-030-00934-2_85

M3 - Conference proceeding

AN - SCOPUS:85054065475

SN - 9783030009335

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

SP - 768

EP - 776

BT - Medical Image Computing and Computer Assisted Intervention – MICCAI 2018 - 21st International Conference, 2018, Proceedings

A2 - Fichtinger, Gabor

A2 - Davatzikos, Christos

A2 - Alberola-López, Carlos

A2 - Frangi, Alejandro F.

A2 - Schnabel, Julia A.

T2 - 21st International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2018

Y2 - 16 September 2018 through 20 September 2018

ER -

Bortsova G, Dubost F, Ørting S, Katramados I, Hogeweg L, Thomsen L et al. Deep learning from label proportions for emphysema quantification. In Fichtinger G, Davatzikos C, Alberola-López C, Frangi AF, Schnabel JA, editors, Medical Image Computing and Computer Assisted Intervention – MICCAI 2018 - 21st International Conference, 2018, Proceedings. 2018. p. 768-776. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol. 11071 LNCS). doi: 10.1007/978-3-030-00934-2_85

Deep learning from label proportions for emphysema quantification

Abstract

Publication series

Conference

Bibliographical note

UN SDGs

Access to Document

Other files and links

Fingerprint

Cite this