Radiologists may misidentify AI images

CT and MRI images were harder to distinguish than radiographs and ultrasound, while subspecialty expertise was associated with greater accuracy.

1

Radiologists identified 75% of AI-generated images correctly, with accuracy varying by imaging modality and subspecialty expertise.
2

The study involved 182 radiologists reviewing 30 images, including 20 AI-generated and 10 real images across various modalities.
3

CT images had the lowest median accuracy at 70%, while radiographs had the highest accuracy at 91% for identifying real and AI-generated images.
4

Radiologists with relevant subspecialty expertise achieved higher classification accuracy, particularly in chest and cardiac imaging.
5

The study noted limitations, including the presentation format of images and potential selection bias among participants.

1