The world is being quietly rearranged by people who write very long documents.


The title they went with Seeing Like Radiologists: Context- and Gaze-Guided Vision-Language Pretraining for Chest X-rays Noisy translates that to

AI radiologist assistant now learns from where doctors actually look


Researchers built a machine-learning system that learns from radiologists' eye gaze and clinical context (patient history, symptoms) when analyzing chest X-rays, rather than treating images as isolated pictures. This makes the AI better at spotting disease patterns and explaining its reasoning in medical reports — roughly 2% to 23% more accurate depending on the task.
Medical AI has mostly ignored how radiologists actually work: they read a patient's chart first, then look at specific regions of an image based on what they're searching for. This paper shows that when AI learns this workflow — not just pattern-matching on pixels — it gets measurably better at the actual job radiologists do, which matters because accurate medical AI requires understanding human diagnostic reasoning, not just optimizing benchmarks.

If you insist
Read the original →