Tag
This paper adapts a diffusion language model for interactive radiology report drafting, showing it matches autoregressive models in accuracy while offering unique infill capabilities that allow radiologists to fix report fragments and have the model fill in the text between them.