Harnessing Generative Adversarial Network (GAN) to Generate Synthetic Mitosis Images for Classification of Cell Images

Background: The task of detecting and tracking of mitosis is important in many biomedical areas such as cancer and stem cell research. This task becomes complex when done in a high density cell array, largely due to an extremely unbalanced data, with a very small number of proliferating cell in each image.

Methods: Using the fact that before proliferating, cells seems to get rounder and brighter, our group extracted bright blobs in each image and considered the patch around each blob as a candidate for mitosis. These candidates were labeled and divided into training, validation and test sets, and used for training of a Convolutional Neural Network (CNN). In current work, we wish to augment the performance by generating synthetic patches of mitosis using a GAN. Specifically, we focus on using the positive examples (mitosis patches) of the training set.

Results: Trying to predict the labels of the test set candidates using a CNN trained by both real and the synthetically generated images showed an increase in both sensitivity and specificity, in comparison to a CNN trained only on real examples.

Conclusion: In this study we showed that generating synthetic cell images can assist in training a CNN in a greatly unbalanced environment.