Enhancing Breast Cancer Detection Systems: Augmenting Mammogram Images Using Generative Adversarial Networks

This research delves into the utilization of generative adversarial networks (GANs) for the augmentation of mammogram images extracted from publicly accessible breast cancer datasets like CBIS-DDSM. The primary objective is to produce a more varied set of images to enrich the training data for breast cancer detection systems by training a neural network to learn the inherent characteristics of mammogram images and generate new images which did not exist in the training data. Through a comprehensive analysis, the study evaluates the efficacy of pre-existing methodologies both pre-augmentation and post-augmentation, seeking to ascertain whether an improvement in accuracy can be achieved. The scarcity of available breast cancer datasets, attributed to the labor-intensive curation and labeling of images, coupled with privacy concerns, serves as a driving force behind investigating GANs as a potential solution. This exploration aims to address the challenge of obtaining a more extensive and diverse dataset, essential for the robust training of breast cancer detection systems.