GP-Unet: Lesion detection from weak labels with a 3D regression network

Florian Dubost; Gerda Bortsova; Hieab Adams; Arfan Ikram; Wiro J. Niessen; Meike Vernooij; Marleen de Bruijne

doi:10.1007/978-3-319-66179-7_25

GP-Unet: Lesion detection from weak labels with a 3D regression network

Florian Dubost^*, Gerda Bortsova, Hieab Adams, Arfan Ikram, Wiro J. Niessen, Meike Vernooij, Marleen de Bruijne

^*Corresponding author for this work

Department of Computer Science

23 Citations (Scopus)

Abstract

We propose a novel convolutional neural network for lesion detection from weak labels. Only a single, global label per image - the lesion count - is needed for training. We train a regression network with a fully convolutional architecture combined with a global pooling layer to aggregate the 3D output into a scalar indicating the lesion count. When testing on unseen images, we first run the network to estimate the number of lesions. Then we remove the global pooling layer to compute localization maps of the size of the input image. We evaluate the proposed network on the detection of enlarged perivascular spaces in the basal ganglia in MRI. Our method achieves a sensitivity of 62% with on average 1.5 false positives per image. Compared with four other approaches based on intensity thresholding, saliency and class maps, our method has a 20% higher sensitivity.

Original language	English
Title of host publication	Medical Image Computing and Computer-Assisted Intervention − MICCAI 2017 : 20th International Conference, Quebec City, QC, Canada, September 11-13, 2017, Proceedings, Part III
Editors	Maxime Descoteaux, Lena Maier-Hein, Alfred Franz, Pierre Jannin, D. Louis Collins, Simon Duschesne
Number of pages	8
Publisher	Springer
Publication date	2017
Pages	214-221
ISBN (Print)	978-3-319-66178-0
ISBN (Electronic)	978-3-319-66179-7
DOIs	https://doi.org/10.1007/978-3-319-66179-7_25
Publication status	Published - 2017
Event	20th International Conference on Medical Image Computing and Computer-Assisted Intervention - Quebec City, Canada Duration: 11 Sept 2017 → 13 Sept 2017 Conference number: 20

Conference

Conference	20th International Conference on Medical Image Computing and Computer-Assisted Intervention
Number	20
Country/Territory	Canada
City	Quebec City
Period	11/09/2017 → 13/09/2017

Series	Lecture notes in computer science
Volume	10435
ISSN	0302-9743

Access to Document

10.1007/978-3-319-66179-7_25

https://arxiv.org/pdf/1705.07999.pdf

Cite this

Dubost, F., Bortsova, G., Adams, H., Ikram, A., Niessen, W. J., Vernooij, M., & de Bruijne, M. (2017). GP-Unet: Lesion detection from weak labels with a 3D regression network. In M. Descoteaux, L. Maier-Hein, A. Franz, P. Jannin, D. L. Collins, & S. Duschesne (Eds.), Medical Image Computing and Computer-Assisted Intervention − MICCAI 2017: 20th International Conference, Quebec City, QC, Canada, September 11-13, 2017, Proceedings, Part III (pp. 214-221). Springer. Lecture notes in computer science Vol. 10435 https://doi.org/10.1007/978-3-319-66179-7_25

GP-Unet : Lesion detection from weak labels with a 3D regression network. / Dubost, Florian; Bortsova, Gerda; Adams, Hieab et al.

Medical Image Computing and Computer-Assisted Intervention − MICCAI 2017: 20th International Conference, Quebec City, QC, Canada, September 11-13, 2017, Proceedings, Part III. ed. / Maxime Descoteaux; Lena Maier-Hein; Alfred Franz; Pierre Jannin; D. Louis Collins; Simon Duschesne. Springer, 2017. p. 214-221 (Lecture notes in computer science, Vol. 10435).

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

Dubost, F, Bortsova, G, Adams, H, Ikram, A, Niessen, WJ, Vernooij, M & de Bruijne, M 2017, GP-Unet: Lesion detection from weak labels with a 3D regression network. in M Descoteaux, L Maier-Hein, A Franz, P Jannin, DL Collins & S Duschesne (eds), Medical Image Computing and Computer-Assisted Intervention − MICCAI 2017: 20th International Conference, Quebec City, QC, Canada, September 11-13, 2017, Proceedings, Part III. Springer, Lecture notes in computer science, vol. 10435, pp. 214-221, 20th International Conference on Medical Image Computing and Computer-Assisted Intervention, Quebec City, Canada, 11/09/2017. https://doi.org/10.1007/978-3-319-66179-7_25

Dubost F, Bortsova G, Adams H, Ikram A, Niessen WJ, Vernooij M et al. GP-Unet: Lesion detection from weak labels with a 3D regression network. In Descoteaux M, Maier-Hein L, Franz A, Jannin P, Collins DL, Duschesne S, editors, Medical Image Computing and Computer-Assisted Intervention − MICCAI 2017: 20th International Conference, Quebec City, QC, Canada, September 11-13, 2017, Proceedings, Part III. Springer. 2017. p. 214-221. (Lecture notes in computer science, Vol. 10435). doi: 10.1007/978-3-319-66179-7_25

Dubost, Florian ; Bortsova, Gerda ; Adams, Hieab et al. / GP-Unet : Lesion detection from weak labels with a 3D regression network. Medical Image Computing and Computer-Assisted Intervention − MICCAI 2017: 20th International Conference, Quebec City, QC, Canada, September 11-13, 2017, Proceedings, Part III. editor / Maxime Descoteaux ; Lena Maier-Hein ; Alfred Franz ; Pierre Jannin ; D. Louis Collins ; Simon Duschesne. Springer, 2017. pp. 214-221 (Lecture notes in computer science, Vol. 10435).

@inproceedings{4b619c6fadb0473e98923e75c52684c4,

title = "GP-Unet: Lesion detection from weak labels with a 3D regression network",

abstract = "We propose a novel convolutional neural network for lesion detection from weak labels. Only a single, global label per image - the lesion count - is needed for training. We train a regression network with a fully convolutional architecture combined with a global pooling layer to aggregate the 3D output into a scalar indicating the lesion count. When testing on unseen images, we first run the network to estimate the number of lesions. Then we remove the global pooling layer to compute localization maps of the size of the input image. We evaluate the proposed network on the detection of enlarged perivascular spaces in the basal ganglia in MRI. Our method achieves a sensitivity of 62% with on average 1.5 false positives per image. Compared with four other approaches based on intensity thresholding, saliency and class maps, our method has a 20% higher sensitivity.",

author = "Florian Dubost and Gerda Bortsova and Hieab Adams and Arfan Ikram and Niessen, {Wiro J.} and Meike Vernooij and {de Bruijne}, Marleen",

year = "2017",

doi = "10.1007/978-3-319-66179-7_25",

language = "English",

isbn = "978-3-319-66178-0",

series = "Lecture notes in computer science",

publisher = "Springer",

pages = "214--221",

editor = "Maxime Descoteaux and Lena Maier-Hein and Alfred Franz and Pierre Jannin and Collins, {D. Louis} and Simon Duschesne",

booktitle = "Medical Image Computing and Computer-Assisted Intervention − MICCAI 2017",

note = "20th International Conference on Medical Image Computing and Computer-Assisted Intervention ; Conference date: 11-09-2017 Through 13-09-2017",

}

TY - GEN

T1 - GP-Unet

T2 - 20th International Conference on Medical Image Computing and Computer-Assisted Intervention

AU - Dubost, Florian

AU - Bortsova, Gerda

AU - Adams, Hieab

AU - Ikram, Arfan

AU - Niessen, Wiro J.

AU - Vernooij, Meike

AU - de Bruijne, Marleen

N1 - Conference code: 20

PY - 2017

Y1 - 2017

N2 - We propose a novel convolutional neural network for lesion detection from weak labels. Only a single, global label per image - the lesion count - is needed for training. We train a regression network with a fully convolutional architecture combined with a global pooling layer to aggregate the 3D output into a scalar indicating the lesion count. When testing on unseen images, we first run the network to estimate the number of lesions. Then we remove the global pooling layer to compute localization maps of the size of the input image. We evaluate the proposed network on the detection of enlarged perivascular spaces in the basal ganglia in MRI. Our method achieves a sensitivity of 62% with on average 1.5 false positives per image. Compared with four other approaches based on intensity thresholding, saliency and class maps, our method has a 20% higher sensitivity.

AB - We propose a novel convolutional neural network for lesion detection from weak labels. Only a single, global label per image - the lesion count - is needed for training. We train a regression network with a fully convolutional architecture combined with a global pooling layer to aggregate the 3D output into a scalar indicating the lesion count. When testing on unseen images, we first run the network to estimate the number of lesions. Then we remove the global pooling layer to compute localization maps of the size of the input image. We evaluate the proposed network on the detection of enlarged perivascular spaces in the basal ganglia in MRI. Our method achieves a sensitivity of 62% with on average 1.5 false positives per image. Compared with four other approaches based on intensity thresholding, saliency and class maps, our method has a 20% higher sensitivity.

UR - http://www.scopus.com/inward/record.url?scp=85029548373&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-66179-7_25

DO - 10.1007/978-3-319-66179-7_25

M3 - Article in proceedings

AN - SCOPUS:85029548373

SN - 978-3-319-66178-0

T3 - Lecture notes in computer science

SP - 214

EP - 221

BT - Medical Image Computing and Computer-Assisted Intervention − MICCAI 2017

A2 - Descoteaux, Maxime

A2 - Maier-Hein, Lena

A2 - Franz, Alfred

A2 - Jannin, Pierre

A2 - Collins, D. Louis

A2 - Duschesne, Simon

PB - Springer

Y2 - 11 September 2017 through 13 September 2017

ER -

GP-Unet: Lesion detection from weak labels with a 3D regression network

Abstract

Conference

Access to Document

Other files and links

Fingerprint

Cite this