Generative Adversarial Network as Data Balance and Augmentation Tool in Histopathology of Breast Cancer

Abstract

This paper explores Wasserstein Generative Adversarial Network Gradient-Penalty (WGAN-GP) for data balance in the medical domain where data scarcity and imbalance are common. The study applies Transfer Learning with pre-trained models from ImageNet on histopathological breast cancer data, both unbalanced and balanced. WGAN-GP was used to overcome the challenge of generating synthetic images to balance the data and improve the accuracy of the classification task. The highest results were shown by VGG16 with a balanced dataset by WGAN-GP in 300 epochs (95.40% accuracy, 96.56% sensitivity, 94.91% specificity). Results showed an improvement in accuracy from 84.25% to 95.40%.