Scaling Generative Foundation Models for Chest Radiography with Rectified Flow Transformers

2026-06-17 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Computer Vision · Depth: Expert, quick

Summary

A new generative foundation model for chest radiograph synthesis, named Rectified Flow Transformers, has been introduced, marking the first such model trained from scratch at the billion-parameter scale. This specialist model features over 1.3 billion parameters and was trained on 1.6 trillion tokens using a curated dataset of 1.2 million radiographs with clinical expert-guided metadata. It addresses the generalization issues of existing radiographic AI models across patient subpopulations and acquisition settings. The model supports controllable radiograph generation and editing, allowing for variations across multiple demographic subgroups, acquisition views, and a dozen pathologies. Its synthesis fidelity significantly advances the state of the art, producing images indistinguishable from real radiographs to clinical experts, offering a promising path for diversifying clinical datasets and evaluating diagnostic model robustness.

Key takeaway

For AI Scientists and Research Scientists developing medical imaging models, this advancement in generative foundation models provides a critical tool. You should consider integrating large-scale generative models like Rectified Flow Transformers to overcome generalization issues in diagnostic AI. This approach enables creating diverse, high-fidelity synthetic datasets. These are crucial for robust model training and evaluation across varied patient demographics and pathologies. Utilize this capability to enhance the reliability and clinical utility of your diagnostic AI systems.

Key insights

A 1.3B-parameter generative foundation model for chest radiographs achieves expert-indistinguishable synthesis and controllable editing.

Principles

• Large-scale generative models improve generalization.
• High-fidelity synthesis diversifies clinical datasets.
• Controllable generation enhances diagnostic model evaluation.

Method

The Rectified Flow Transformer was trained from scratch at 1.3B parameters using 1.6T tokens from 1.2M curated, heterogeneous radiographs with expert-guided metadata for controllable synthesis.

In practice

• Generate diverse clinical datasets.
• Evaluate diagnostic model robustness.
• Edit radiographs for specific pathologies.

Topics

• Generative Foundation Models
• Chest Radiography
• Rectified Flow Transformers
• Medical Imaging AI
• Data Augmentation
• Image Synthesis

Best for: AI Scientist, Research Scientist, Computer Vision Engineer

Related on AIssential

• Synthesize Realistic 3D Medical Images at Scale to Ship Pre‑Trained Models — NVIDIA's MAISI and NV-Generate frameworks provide scalable, open-source 3D medical image synthesis to overcome data bottlenecks.
• Scaling Recurrence-aware Foundation Models for Clinical Records via Next-Visit Prediction — RAVEN is a generative pretraining strategy for EHRs that predicts next-visit events while accounting for recurrence.
• Simulating clinical interventions with a generative multimodal model of human physiology — HealthFormer is a generative transformer modeling human physiology for forecasting, risk stratification, and intervention simulation.
• RA-CMF: Region-Adaptive Conditional MeanFlow for CT Image Reconstruction — RA-CMF enhances CT image quality and radiomic consistency by adaptively refining heterogeneous regions using conditional MeanFlow and reinforcement learning.
• When is 3D Worth It? A Resource-Performance Frontier for CNNs and Transformers in Lung CT — Input dimensionality significantly impacts model performance and stability in volumetric medical imaging.
• Impact of Synthetic Lesional MR Images in Automated Focal Cortical Dysplasia Detection in Low-Data Scenarios — Synthetic FCD-MRIs generated by conditional networks can augment real data, reducing labeled data needs for automated detection.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.