Modern databases of biomedical images have been investigated. Biomedical imaging has been shown to be expensive and time consuming. A database of images of precancerous and cancerous breasts "BPCI2100" was developed. The database consists of 2,100 image files and a MySQL database of medical research information (patient information and image features). Generative adversarial networks (GAN) have been found to be an effective means of image generation. The architecture of the generative adversarial network consisting of a generator and a discriminator has been developed.The discriminator is a deep convolutional neural network with color images of 128×128 pixels. This network consists of six convolutional layers with a window size of 5×5 pixels. Leaky ReLU type activation function for convolutional layers is used. The last layer used a sigmoid activation function. The generator is a neural network consisting of a fully connected layer and seven deconvolution layers with a 5×5 pixel window size. Leaky ReLU activation function is used for all layers. The last layer uses the hyperbolic tangent activation function. Google Cloud Compute Instance tools have been used to train the the generative adversarial network. Generation of histological and cytological images on the basis of the generative adversarial network is conducted. As a result, the training sample for classifiers has been significantly increased.
Original histological images are divided into 5 classes, cytological images into 4 classes. The original sample size for histological images is 91 images, for cytological images – 78 images. Training samples have been expanded to 1000 images by affine transformations (shift, zoom, rotate, reflection). Studying the classifier on the original sample yielded an accuracy of ≈84 % for histological images and ≈75 % for cytological images, respectively. On the sample of the generated images, the initial classification accuracy was ≈96.5 % for histological images and ≈95.5 for cytological images. The accuracy gain is ≈12 % for histological images and ≈20.5 % for cytological images. The performed classification of histological and cytological images showed that the increase in classification accuracy was ≈12 % for histological images and ≈20.5 % for cytological images. Computer experiments have shown that the time of study of the generative adversarial network for histological images was ≈9 hours, for cytology – ≈ 8.5 hours. Prospects for further research are the parallelization of algorithms for training generative-competitive networks.
[1] American Society. (2019). Online atlas of the American Society for Cytopathology. Retrieved from: https://bethesda.soc.wisc.edu/. (Date of appeal: September 2019). [In Ukrainian].
[2] Benchmark dataset. (2019). UCSB Bio-Segmentation Benchmark dataset for testing computer vision algorithms, including segmentation and classification. Retrieved from: https://bioimage.ucsb.edu/research/bio-segmentation. (Date of appeal: September 2019). [In Ukrainian].
[3] Berezsky, O. M., Batko, Y. M., Berezka, K. M., Verbovyi, S. O., et al. (2017). Methods, algorithms and software for processing biomedical images. Ternopil: Economic Thought, TNEU, 350 p. [In Ukrainian].
[4] Berezsky, O. M., Melnyk, G. M., Nikolyuk, V. D., & Datsko, T. V. (2013). Database of digital histological and cytological images of various forms of breast cancer "CIFDB": Certificate of copyright registration to the work № 52743 of 23.12.2013. [In Ukrainian].
[5] Creswell, Antonia, White, Tom, Dumoulin, Vincent, Arulkumaran, Kai, Sengupta, Biswa, & Bharath, Anil A. (2018). Generative Adversarial Networks, an overview, (pp. 53–65). https://doi.org/10.1109/MSP.2017.2765202.
[6] Fine-needle Aspirate. (2019). Breast Cancer Wisconsin (Diagnostic) Data Set. The signs were obtained from digitized images of the Fine-needle Aspirate (FNA) breast tissue. The features reflect the characteristics of the cell nuclei. Retrieved from: http://archive.ics.uci.edu/ml/datasets/Breast+Cancer+Wisconsin+%28Diagnostic29 (Date of appeal: September 2019). [In Ukrainian].
[7] Frid-Adar, M., Idit Diamant, Eyal Khang, Michal Amitai, Jacob Goldberger and Hayit Greenspan. (2018). GAN-based Synthetic Medical Image Augmentation for increased CNN Performance in Liver Lesion Classification, (pp. 321–331). https://doi.org/10.1016/j.neucom.2018.09.013
[8] Guibas, John T., Virdi, Tejpal S., & Li, Peter S. (2017). Synthetic Medical Images from Dual Generative Adversarial Networks. 31st Conference on Neural Information Processing Systems (NIPS2017) and LongBeach and CA and USA, (p. 9).
[9] He Zhao, Huiqi Li, Sebastian Maurer-Stroh, & Li Cheng. (2018). Synthesizing Retinal and Neuronal Images with Generative. Adversarial Nets, journal = {Medical Image Analysis}, chapter= {49}, S1361-8415(18)30459-6. , (pp. 14–26). https://doi.org/10.1016/j.media.2018.07.001.
[10] Histology Guide. (2019). Virtual histology laboratory: Histology Guide. Retrieved from: http://www.histologyguide.com/index.html (Date of appeal: September 2019). [In Ukrainian].
[11] Kansas School. (2019). Histological online atlas for laboratory and cell biology courses at the University of Kansas School of Medicine. Retrieved from: http://www.kumc.edu/instruction/medicine/anatomy/hisoweb/index.html (Date of appeal: September 2019). [In Ukrainian].
[12] Kazeminiaa, Salome, Baurb, Christoph, Kuijperc, Arjan, Innekend, Bramvan, Navabb, Nassir, Albarqouni, Shadi, Mukhopadhyaya, Anirban. (2018). GANs for Medical Image Analysis", journal = {Department of Computer Science and TU Darmstadt and Germany Computer Aided Medical Procedures (CAMP) and TU Munich Germany}, (pp. 1–41).
[13] Lucas Theis. (2016). Lucas Theis Aaron "van den Oord y University of Tubingen" Ghent University 72072 Tubingen, Matthias Bethge, "A Note On The Evaluation Of Generative Models", Published as a conference paper at ICLR 2016, arXiv:1511.01844v3 [stat. ML] 24 Apr 2016, p. 10.
[14] Salimans, Tim, Goodfellow, Ian, Zaremba, Wojciech, Cheung, Vicki, Radford, Alec, & Chen, Xi. (2016). Improved Techniques for Training GANs. In arXiv, (pp. 2234–2242).
[15] Xian Wu, Kun Xu, & Peter Hall. (2017). A Survey of Image Synthesis and Editing with Generative Adversarial Networks. Tsinghua science and technology, 15. ISSNll1007-0214ll09/15llpp660-674.