CheXGenBench: A Unified Benchmark For Fidelity, Privacy and Utility of Synthetic Chest Radiographs
Researchers have introduced CheXGenBench, the first unified evaluation framework for synthetic chest radiograph generation. This benchmark assesses generative fidelity, privacy risks, and downstream utility across various text-to-image models. The study found that current models struggle with long-tailed medical distributions, pose significant privacy risks, and have limited utility for multimodal tasks, despite benefiting downstream classification. AI
IMPACT Establishes a new standard for evaluating synthetic medical data, potentially guiding future development of more robust and privacy-preserving generative models.