AI model predicts brain tumour MRI contrast enhancement, reducing need for gadolinium injections

UCL researchers developed an artificial-intelligence algorithm that predicts the appearance of brain tumours on contrast-enhanced MRI using only non-contrast images. The model was trained with 11,089 scans from more than 8,500 patients in four countries and tested on over 1,100 non-contrast studies, achieving an overall correct prediction rate of 83%. It identified 92% of tumours that would brighten with gadolinium and 74% of those that would not, performing best on meningioma and less well in children.

While not yet ready to replace contrast-enhanced MRI, the system could act as a triage tool, indicating when contrast might be necessary and reducing unnecessary dye exposure and follow-up appointments. Experts highlighted unknown long-term health effects of gadolinium and its environmental contamination as additional motivations. Dr Karen Noble of Brain Tumour Research welcomed the innovation as a potential way to improve patient care and limit side effects.