Biomedical Data Augmentation

Introduction

Synthesizing photo-realistic images is a challenging problem with many practical applications. In many cases, the availability of a significant amount of images is crucial, yet obtaining them might be not trivial. For instance, obtaining huge databases of images is hard, in the biomedical domain, but strictly needed in order to improve both algorithms and physicians’ skills. In the latest years, new deep learning mod- els have been proposed in the literature, called Generative Adversarial Neural Networks (GANNs), that turned out as effective at synthesizing high-quality image in several domains.

Methods

Generative Adversarial Neural Network is a generative model approach based on differentiable generator networks. GANNs are conceived for scenarios in which the generator network must compete against an adversary, in a sort of forger-police relation.

Two actors are involved: the Generator network (the “forger”), which directly produces samples that should look as they came from a particular domain (e.g. the training set) and the Discriminator (the “police”), that attempts to distinguish between samples taken from the original data and samples drawn from the Generator. 

Interestingly, this approach (and its variants) can be used to actually generate unseen images, including medical images.  And this is what we did and what we do in this research project. 

Look below, can you tell which is fake?

Publications

[1] F. Calimeri, A. Marzullo, C. Stamile, and G. Terracina, "Biomedical data augmentation using generative adversarial neural networks", in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10614 LNCS, pp. 626-634, 2017. 26th International Conference on Artificial Neural Networks, ICANN 2017, 11-14 September 2017. DOI: 10.1007/978-3-319-68612-7_71

[2] A. Marzullo, S. Moccia, M. Catellani, F. Calimeri, and E. D. Momi, "Towards realistic laparoscopic image generation using image-domain translation", Computer Methods and Programs in Biomedicine, vol. 200, no. 105834, 2021. DOI: 10.1016/j.cmpb.2020.105834