Abstract
Prostate cancer is a leading cause of cancer-related death in men globally. Deep learning algorithms hold significant promise for disease classification and detection, but their performance heavily relies on the availability of large training datasets. Generative Adversarial Networks (GANs) and Diffusion Models offer cutting-edge solutions to address this challenge by synthesizing realistic medical data.
This work investigates the potential of StyleGAN2-ADA and Latent Diffusion Models (LDMs) for generating synthetic T2-weighted prostate MRI slices. We leverage a publicly available dataset to train these models and evaluate their ability to create realistic MRI data. Our initial findings demonstrate that both frameworks can produce slices visually similar to real data. However, a crucial distinction lies in their training and inference efficiency.
