Scaling Generative Foundation Models for Chest Radiography with Rectified Flow Transformers

The article presents a new generative foundation model for chest radiograph synthesis, boasting over 1.3 billion parameters and trained on 1.6 trillion tokens from a diverse dataset of 1.2 million radiographs. This model enhances the fidelity of synthesized images, achieving results indistinguishable from real radiographs, and supports controlled generation across various demographic subgroups and pathologies. This advancement is significant for practitioners as it addresses the limitations of existing models in generalization and clinical applicability, facilitating the creation of more robust diagnostic tools.