Artificial intelligence is no longer just helping interpret medical images. It can now create highly convincing fake ones. According to a new study in Radiology, those synthetic radiographs are realistic enough to fool both radiologists and large language models (LLMs). In this retrospective diagnostic accuracy study, investigators examined whether radiologists and multimodal AI systems could distinguish authentic X-rays from synthetic ones generated by GPT-4o and RoentGen, a chest X-ray diffusion model. The concern goes beyond technological novelty.
It includes the growing potential for misuse in clinical care, research datasets, training pipelines, and even medicolegal settings. The study included 17 radiologists from six countries, with experience ranging from residency to 40 years in practice. In the first phase, readers were not told that fake images were included. Among 154…
