Generating High-quality Synthetic Mammogram Images Using Denoising Diffusion Probabilistic Models: a Novel Approach for Augmenting Deep Learning Datasets

Deep learning models for breast cancer detection require large and diverse datasets of mammogram images to achieve high accuracy. However, the availability of such datasets is limited due to privacy concerns, high annotation costs, and the rarity of certain pathological cases. Traditional data augmentation methods like flipping, rotation, and cropping can enhance dataset size and variation but cannot generate new realistic pathological conditions. Advanced generative artificial intelligence techniques can produce synthetic images, including medical imaging. This study proposes a DDPM-based framework for generating high-quality synthetic mammogram images to augment deep learning datasets, demonstrating superior performance compared to traditional and contemporary augmentation methods. The generated images are evaluated using Fréchet Inception Distance (FID), Precision, and Recall metrics, highlighting the potential of DDPMs to enhance breast cancer detection models using deep-learning methods.