A new AI tool could help predict which brain tumours will show up on MRI scans without the need for contrast fluid injections, finds a study led by UCL researchers and supported by the BRC.
The tool could help doctors determine which patients can safely skip the injection, sparing some people an unnecessary dose of contrast dye, according to the findings published in The Lancet Medical Imaging & Theranostics.
Doctors routinely inject a contrast dye called gadolinium during brain MRI scans to make growing or aggressive tumour tissue light up, which helps guide diagnosis and treatment. However, gadolinium can build up in the body, and its long-term effects are not yet fully understood. In rare cases, it has been linked to a serious complication in people with severe kidney disease. It is generally avoided in young children and in people with kidney problems or allergies, and used sparingly when repeated scans are needed. Gadolinium also raises potential environmental concerns because it does not fully break down, and traces have been found in sewage, rivers, and drinking water.
Using more than 11,000 MRI scans from over 8,500 patients across four countries, the researchers developed and tested an AI tool to predict exactly which parts of the brain would enhance if contrast fluid had been given. The dataset included adults and children, several tumour types, and scans taken on different machines between 2006 and 2024 to reflect real-world diversity.
On a set of 1,109 non-contrast images, the tool correctly identified 92% of the tumours that would brighten with contrast dye and 74% of those that would not. Accuracy varied by tumour type and patient group. It was most accurate for meningioma, a common brain tumour (93 of 100 cases), and least accurate for children (9 of 20 cases), likely because children were underrepresented in the training data.
The authors say the findings point to a promising new way to detect brain tumours and could eventually reduce gadolinium use where appropriate. They caution, however, that the tool is not accurate enough to replace contrast-enhanced MRI, which remains the most reliable way to confirm whether a tumour is growing or behaving aggressively.
Lead author Dr James Ruffle (UCL Queen Square Institute of Neurology and National Hospital for Neurology and Neurosurgery, UCLH) said: “Deciding whether a patient needs a contrast injection is a common judgement in brain tumour imaging, yet it is not always straightforward, or even possible. We have shown that artificial intelligence can read the subtle clues hidden in an ordinary scan, many of which expert radiologists find hard to see, and use them to help decide whether a patient needs contrast. Our hope is to make imaging safer for each individual affected, while reducing our dependence on gadolinium.”
The team’s work was funded by the European Society of Radiology, the European Institute for Biomedical Imaging Research, the Medical Research Council, Wellcome, and the NIHR Biomedical Research Centre: UCLH.
Image credit: Dr James Ruffle and Prof Parashkev Nachev
