DermGAN: Synthetic Generation of Clinical Skin Images with Pathology

Amirata Ghorbani, Vivek Natarajan, David Coz, Yuan Liu
; Proceedings of the Machine Learning for Health NeurIPS Workshop, PMLR 116:155-170, 2020.

Abstract

Despite the recent success in applying supervised deep learning to medical imaging tasks, the problem of obtaining large and diverse expert-annotated datasets required for the development of high performant models remains particularly challenging. In this work, we explore the possibility of using Generative Adverserial Networks (GAN) to synthesize clinical images with skin condition. We propose DermGAN, an adaptation of the popular Pix2Pix architecture, to create synthetic images for a pre-specified skin condition while being able to vary its size, location and the underlying skin color. We demonstrate that the generated images are of high fidelity using objective GAN evaluation metrics. In a Human Turing test, we note that the synthetic images are not only visually similar to real images, but also embody the respective skin condition in dermatologists’ eyes. Finally, when using the synthetic images as a data augmentation technique for training a skin condition classifier, we observe that the model performs comparably to the baseline model overall while improving on rare but malignant conditions.

Cite this Paper


BibTeX
@InProceedings{pmlr-v116-ghorbani20a, title = {{DermGAN: Synthetic Generation of Clinical Skin Images with Pathology}}, author = {Ghorbani, Amirata and Natarajan, Vivek and Coz, David and Liu, Yuan}, booktitle = {Proceedings of the Machine Learning for Health NeurIPS Workshop}, pages = {155--170}, year = {2020}, editor = {Adrian V. Dalca and Matthew B.A. McDermott and Emily Alsentzer and Samuel G. Finlayson and Michael Oberst and Fabian Falck and Brett Beaulieu-Jones}, volume = {116}, series = {Proceedings of Machine Learning Research}, address = {}, month = {13 Dec}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v116/ghorbani20a/ghorbani20a.pdf}, url = {http://proceedings.mlr.press/v116/ghorbani20a.html}, abstract = {Despite the recent success in applying supervised deep learning to medical imaging tasks, the problem of obtaining large and diverse expert-annotated datasets required for the development of high performant models remains particularly challenging. In this work, we explore the possibility of using Generative Adverserial Networks (GAN) to synthesize clinical images with skin condition. We propose DermGAN, an adaptation of the popular Pix2Pix architecture, to create synthetic images for a pre-specified skin condition while being able to vary its size, location and the underlying skin color. We demonstrate that the generated images are of high fidelity using objective GAN evaluation metrics. In a Human Turing test, we note that the synthetic images are not only visually similar to real images, but also embody the respective skin condition in dermatologists’ eyes. Finally, when using the synthetic images as a data augmentation technique for training a skin condition classifier, we observe that the model performs comparably to the baseline model overall while improving on rare but malignant conditions.} }
Endnote
%0 Conference Paper %T DermGAN: Synthetic Generation of Clinical Skin Images with Pathology %A Amirata Ghorbani %A Vivek Natarajan %A David Coz %A Yuan Liu %B Proceedings of the Machine Learning for Health NeurIPS Workshop %C Proceedings of Machine Learning Research %D 2020 %E Adrian V. Dalca %E Matthew B.A. McDermott %E Emily Alsentzer %E Samuel G. Finlayson %E Michael Oberst %E Fabian Falck %E Brett Beaulieu-Jones %F pmlr-v116-ghorbani20a %I PMLR %J Proceedings of Machine Learning Research %P 155--170 %U http://proceedings.mlr.press %V 116 %W PMLR %X Despite the recent success in applying supervised deep learning to medical imaging tasks, the problem of obtaining large and diverse expert-annotated datasets required for the development of high performant models remains particularly challenging. In this work, we explore the possibility of using Generative Adverserial Networks (GAN) to synthesize clinical images with skin condition. We propose DermGAN, an adaptation of the popular Pix2Pix architecture, to create synthetic images for a pre-specified skin condition while being able to vary its size, location and the underlying skin color. We demonstrate that the generated images are of high fidelity using objective GAN evaluation metrics. In a Human Turing test, we note that the synthetic images are not only visually similar to real images, but also embody the respective skin condition in dermatologists’ eyes. Finally, when using the synthetic images as a data augmentation technique for training a skin condition classifier, we observe that the model performs comparably to the baseline model overall while improving on rare but malignant conditions.
APA
Ghorbani, A., Natarajan, V., Coz, D. & Liu, Y.. (2020). DermGAN: Synthetic Generation of Clinical Skin Images with Pathology. Proceedings of the Machine Learning for Health NeurIPS Workshop, in PMLR 116:155-170

Related Material