The Role of Synthetic Data and Generative AI in Breast Imaging: Promise, Pitfalls, and Pathways Forward

Abstract

Artificial intelligence is reshaping breast imaging, yet progress is constrained by data scarcity, privacy restrictions, and uneven representation. This narrative review synthesizes evidence (2020–April 2025) on synthetic data and generative AI—principally GANs and diffusion models—in mammography and related modalities. We examine how synthetic images enable data augmentation, class balancing, external validation, and simulation-based training; summarize reported gains in detection performance; and assess their potential to mitigate or, if misapplied, amplify bias across subgroups (age, density, ethnicity). We analyze threats to validity, including enriched cohorts, distribution shift, and unverifiable realism, and address medico-legal exposure, image provenance, and deepfake risks. Finally, we outline task-specific validation and reporting practices, equity auditing across density and demographics, and governance pathways aligned with EU/US regulatory expectations. Synthetic data and generative AI can enhance performance, training, and data sharing; however, responsible clinical adoption requires rigorous validation, transparency on failure modes, tamper-evident provenance, and shared accountability models.