GENERATION OF ARTIFICIAL CHEST RADIOGRAPHS USING GENERATIVE ADVERSARIAL NEURAL NETWORKS
Abstract
This paper deals with the problem of generating artificial chest x-ray images which expected to be almost undistinguishable from real ones. Generation was performed using Generative Adversarial Nets (GAN). Similarity of resultant artificial images to the real ones was evaluated both by visual examination and by quantitative assessment using commonly known Local Binary Patterns. It was concluded that GANs can be successfully employed for generating realistically appearing artificial chest radiographs. However, an automatic procedure of selecting “most realistic” results is necessary for excluding the final visual quality control stage and making the whole generation process fully automatic.
About the Authors
V. A. Kovalev
The United Institute of Informatics Problems of the National Academy of Sciences of Belarus, Minsk
Belarus
Belarus
S. A. Kozlovski
The United Institute of Informatics Problems of the National Academy of Sciences of Belarus, Minsk
Belarus
Belarus
A. A. Kalinovsk
The United Institute of Informatics Problems of the National Academy of Sciences of Belarus, Minsk
Belarus
Belarus
References
1. A survey on deep learning in medical image analysis / G. Litjens [et al.] // Medical Image Analysis. – 2017. – Vol. 42. – P. 60–88.
2. A comparison of conventional and Deep Learning methods of image classification on a database of chest radiographs / V. А. Kovalev [et al.] // Intern. J. of Computer Assisted Radiology and Surgery. – 2017. – Vol. 12, suppl. 1. – P. S139–S140.
3. Benchmarking the efficiency of Deep Learning methods on the problem of predicting subjects’ age by chest radiographs / V. А. Kovalev [et al.] // Proc. 3rd Intern. Conf. Big Data and Advanced Analytics. – Minsk : BSUIR, 2017. – P. 75–82.
4. Generative Adversarial Networks / I. J. Goodfellow [et al.] // arXiv.org [Electronic resource]. – 2014. – Mode of access : https://arxiv.org/pdf/1406.2661. – Date of access : 25.09.2017.
5. Auto-encoding variational bayes / D. P. Kingma [et al.] // arXiv.org [Electronic resource]. – 2014. – Mode of access : https://arxiv.org/pdf/1312.6114. – Date of access : 25.09.2017.
6. LeCun, Y. Gradient-based learning applied to document recognition / Y. LeCun // Proc. of the IEEE. – 1998. – Vol. 86(11). – P. 2278–2324.
7. Unsupervised representation learning with deep convolutional generative adversarial networks / A. Radford, L. Metz, S. Chintala // arXiv.org [Electronic resource]. – 2016. – Mode of access : https://arxiv.org/pdf/1511.06434. – Date of access : 25.09.2017.
8. DCGAN in tensorflow [Electronic resource]. – Mode of access : https://github.com/carpedm20/DCGAN-tensorflow. – Date of access : 25.09.2017.
9. Medical image synthesis with context-aware generative adversarial networks / D. Nie [et al.] // MICCAI, LNCS. –Springer, Cham, 2017. – Vol. 10435. – P. 417–425.
10. Deep MR to CT Synthesis using Unpaired Data / J. M. Wolterink [et al.] // arXiv.org [Electronic resource]. – 2017. – Mode of access : https://arxiv.org/pdf/1708.01155. – Date of access : 22.09.2017.
11. Synthesis of positron emission tomography (PET) images via multi-channel Generative Adversarial Networks (GANs) / L. Bi [et al.] // Molecular Imaging, Reconstruction and Analysis of Moving Body Organs, and Stroke Imaging and Treatment. – Springer, Cham, 2017. – Vol. 10555. – P. 43–51.
Review
For citations:
Kovalev V.A., Kozlovski S.A., Kalinovsk A.A. GENERATION OF ARTIFICIAL CHEST RADIOGRAPHS USING GENERATIVE ADVERSARIAL NEURAL NETWORKS. Informatics. 2018;15(2):7-16. (In Russ.)
Views: 1224
Email this article 