Conditional Generative Adversarial Networks for Data Augmentation in Breast Cancer Classification

Abstract

Automatic breast cancer classification benefits pathologists in obtaining fast and precise diagnoses and improving early detection. However, the performance of deep learning models depends greatly on the quality and quantity of the datasets used. Due to the complexity and high costs of patient data collection, many medical datasets, particularly for pathological conditions, suffer from small sample sizes. Hence, developing a deep learning solution for breast cancer classification is still challenging. Data augmentation is one of the popular approaches to bridge this gap. In this work, we propose to use Conditional Generative Adversarial Networks (CGANs) for data augmentation. The aim of training CGANs is to generate a new set of realistic synthetic images and combine these together with real images to form a new augmented training set. The experiments show that most of the images produced by CGAN are reliable and classification performance with CGAN-based data augmentation can achieve good results. This method, unlike traditional data augmentation, can produce histopathological images that are completely different from the existing data. Therefore, this technique has the potential to address data scarcity and to directly benefit the training of deep learning models.