FleXray: Universal Clinical X-ray Segmentation
FleXray is trained on simulated, fully labeled X-rays rather than a large hand-labeled clinical X-ray set.
The paper presents a generalist segmentation model for clinical X-rays across the whole body. Its training data comes from a physics-based engine that turns existing 3D CT segmentation datasets, plus generative image editing, into annotated 2D X-ray simulations. The authors report segmentation of 60 anatomical structures on unseen research datasets and in-the-wild X-rays. They also say the system supports quantitative X-ray analysis for disease grading, image-guided intervention navigation, and learning pathological targets with less data. ArXiv · AI/CL/LG's note
The paper presents a generalist segmentation model for clinical X-rays across the whole body. Its training data comes from a physics-based engine that turns existing 3D CT segmentation datasets, plus generative image editing, into annotated 2D X-ray simulations. The authors report segmentation of 60 anatomical structures on unseen research datasets and in-the-wild X-rays. They also say the system supports quantitative X-ray analysis for disease grading, image-guided intervention navigation, and learning pathological targets with less data. ArXiv · AI/CL/LG's note
score 6