Frontier generative foundation model for chest radiography. a) The proposed CXR7-1M dataset, harmonised from seven existing datasets and augmented with additional radiologist-guided metadata. b) Radiographic rectified flow transformer (RadiT), and VAE trained with a domain-specific Rad-DINO perceptual loss (Rad-VAE). c) Synthetic 512x512 resolution chest radiographs generated using our RadiT XL (1.3B) model.
We introduce the first generative foundation model for chest radiograph synthesis trained from scratch at the billion-parameter scale. Existing radiographic AI models often suffer from poor generalisation across patient subpopulations, institutions, and acquisition settings, resulting in limited real-world clinical utility. Controlled, high-fidelity synthesis of chest radiographs is a promising path toward diversifying clinical datasets and evaluating the robustness of diagnostic models. Therefore, we present the largest specialist generative foundation model for chest radiographs to date, with over 1.3B parameters, trained for 1.6T tokens on a curated, heterogeneous dataset comprising 1.2M radiographs and clinical expert-guided metadata. Our model supports controllable radiograph generation and editing across multiple demographic subgroups, acquisition views, and a dozen pathologies. Moreover, we significantly advance the state-of-the-art in radiograph synthesis fidelity, producing images that are indistinguishable from real radiographs to clinical experts.
Test yourself on the same real-vs-synthetic challenge used with clinical experts.
Choose the synthetic radiograph generated by RadiT XL. Good luck!
We collate the CXR7-1M dataset, the largest-scale open-source chest X-ray dataset to date, comprising over 1.2M radiographs, harmonised from multiple existing datasets and paired with radiologist-guided metadata systematically extracted through expert consultation.
We build a series of scaled rectified flow transformers for chest X-ray generation, up to 1.3B parameters. Our largest model, RadiT XL, was trained for 1.6T tokens attains four-fold FDD and ten-fold KDD improvements over prior state-of-the-art on the CheXGenBench benchmark.
We introduce an expert-designed, clinically plausible causal graph for chest X-ray and instantiate it as the largest continuous flow-based causal model to date, spanning 19 demographic and radiological variables, and unlocking scalable exact abduction for discrete factors.
We evaluate our models on controllable image generation and editing, showing that RadiT XL achieves high-fidelity control over multiple demographic, acquisition-view, and clinical attributes.
We conduct a blind real-vs-synthetic discrimination experiment and show that RadiT XL produces chest radiographs that expert clinicians find indistinguishable from real ones.
