Red-GAN: Attacking class imbalance via conditioned generation. Yet another medical imaging perspective.

Ahmad B. Qasim, Ivan Ezhov, Suprosanna Shit, Oliver Schoppe, Johannes C. Paetzold, Anjany Sekuboyina, Florian Kofler, Jana Lipkova, Hongwei Li, Bjoern Menze
Proceedings of the Third Conference on Medical Imaging with Deep Learning, PMLR 121:655-668, 2020.

Abstract

Exploiting learning algorithms under scarce data regimes is a limitation and a reality of the medical imaging field. In an attempt to mitigate the problem, we propose a data augmentation protocol based on generative adversarial networks. We condition the networks at a pixel-level (segmentation mask) and at a global-level information (acquisition environment or lesion type). Such conditioning provides immediate access to the image-label pairs while controlling global class specific appearance of the synthesized images. To stimulate synthesis of the features relevant for the segmentation task, an additional passive player in a form of segmentor is introduced into the the adversarial game. We validate the approach on two medical datasets: BraTS, ISIC. By controlling the class distribution through injection of synthetic images into the training set we achieve control over the accuracy levels of the datasets’ classes. The code is available at https://github.com/IvanEz/Red-GAN.

Cite this Paper

BibTeX
@InProceedings{pmlr-v121-qasim20a, title = {Red-GAN: Attacking class imbalance via conditioned generation. Yet another medical imaging perspective.}, author = {Qasim, Ahmad B. and Ezhov, Ivan and Shit, Suprosanna and Schoppe, Oliver and Paetzold, Johannes C. and Sekuboyina, Anjany and Kofler, Florian and Lipkova, Jana and Li, Hongwei and Menze, Bjoern}, booktitle = {Proceedings of the Third Conference on Medical Imaging with Deep Learning}, pages = {655--668}, year = {2020}, editor = {Arbel, Tal and Ben Ayed, Ismail and de Bruijne, Marleen and Descoteaux, Maxime and Lombaert, Herve and Pal, Christopher}, volume = {121}, series = {Proceedings of Machine Learning Research}, month = {06--08 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v121/qasim20a/qasim20a.pdf}, url = {https://proceedings.mlr.press/v121/qasim20a.html}, abstract = {Exploiting learning algorithms under scarce data regimes is a limitation and a reality of the medical imaging field. In an attempt to mitigate the problem, we propose a data augmentation protocol based on generative adversarial networks. We condition the networks at a pixel-level (segmentation mask) and at a global-level information (acquisition environment or lesion type). Such conditioning provides immediate access to the image-label pairs while controlling global class specific appearance of the synthesized images. To stimulate synthesis of the features relevant for the segmentation task, an additional passive player in a form of segmentor is introduced into the the adversarial game. We validate the approach on two medical datasets: BraTS, ISIC. By controlling the class distribution through injection of synthetic images into the training set we achieve control over the accuracy levels of the datasets’ classes. The code is available at https://github.com/IvanEz/Red-GAN.} }
Endnote
%0 Conference Paper %T Red-GAN: Attacking class imbalance via conditioned generation. Yet another medical imaging perspective. %A Ahmad B. Qasim %A Ivan Ezhov %A Suprosanna Shit %A Oliver Schoppe %A Johannes C. Paetzold %A Anjany Sekuboyina %A Florian Kofler %A Jana Lipkova %A Hongwei Li %A Bjoern Menze %B Proceedings of the Third Conference on Medical Imaging with Deep Learning %C Proceedings of Machine Learning Research %D 2020 %E Tal Arbel %E Ismail Ben Ayed %E Marleen de Bruijne %E Maxime Descoteaux %E Herve Lombaert %E Christopher Pal %F pmlr-v121-qasim20a %I PMLR %P 655--668 %U https://proceedings.mlr.press/v121/qasim20a.html %V 121 %X Exploiting learning algorithms under scarce data regimes is a limitation and a reality of the medical imaging field. In an attempt to mitigate the problem, we propose a data augmentation protocol based on generative adversarial networks. We condition the networks at a pixel-level (segmentation mask) and at a global-level information (acquisition environment or lesion type). Such conditioning provides immediate access to the image-label pairs while controlling global class specific appearance of the synthesized images. To stimulate synthesis of the features relevant for the segmentation task, an additional passive player in a form of segmentor is introduced into the the adversarial game. We validate the approach on two medical datasets: BraTS, ISIC. By controlling the class distribution through injection of synthetic images into the training set we achieve control over the accuracy levels of the datasets’ classes. The code is available at https://github.com/IvanEz/Red-GAN.
APA
Qasim, A.B., Ezhov, I., Shit, S., Schoppe, O., Paetzold, J.C., Sekuboyina, A., Kofler, F., Lipkova, J., Li, H. & Menze, B.. (2020). Red-GAN: Attacking class imbalance via conditioned generation. Yet another medical imaging perspective.. Proceedings of the Third Conference on Medical Imaging with Deep Learning, in Proceedings of Machine Learning Research 121:655-668 Available from https://proceedings.mlr.press/v121/qasim20a.html.

Related Material