AI tool may help certain brain tumor patients avoid repeat GBCA exposure
A new artificial intelligence-enabled tool could help eliminate the need for gadolinium-based contrast agents during MRI scans used to monitor brain tumors.
GBCAs are routinely used during brain MRIs to highlight tumor tissue and help clinicians assess disease extent and behavior. Although widely used, gadolinium can accumulate in the body, and its use may be limited in certain patients, such as those with kidney disease or allergies. Repeated contrast-enhanced examinations also can increase cumulative exposure.
Experts at University College London have developed an AI tool to help address this issue. It is intended to help predict which areas of a brain tumor would show enhancement if contrast were administered. Researchers believe the approach could eventually help reduce unnecessary gadolinium exposure for some patients while maintaining useful information for diagnosis and treatment planning.
“Brain tumor imaging assessment typically requires both pre-contrast and post-contrast MRI, but gadolinium administration is not always desirable—for example, in patients requiring frequent follow-up, or those with renal impairment or allergy, or those who are of pediatric age,” lead author James Ruffle, MD, of the UCL Queen Square Institute of Neurology and the National Hospital for Neurology and Neurosurgery at UCLH, and colleagues noted. “We aimed to develop and validate a deep learning model to predict brain tumor contrast enhancement from non-contrast MRI sequences alone.”
For the study, researchers used over 11,000 MRI scans from more than 8,500 patients across four countries and three continents to develop and train their DL model. The dataset included adults and children, multiple tumor types and exams performed on different MRI systems between 2006 and 2024. The AI model was trained to identify imaging features on non-contrast MRI that could predict subsequent contrast enhancement.
On a test set of over 1,100 non-contrast scans, the model correctly predicted whether a tumor would enhance with contrast approximately 83% of the time. Its sensitivity was 92% for singling out tumors that would enhance, while its specificity was 74% for tumors that did not enhance.
Its performance varied among patient groups and different tumor types. The model performed particularly well for meningiomas, correctly classifying 93 of 100 cases. However, its performance was substantially lower in children; it correctly classified nine of 20 cases, which the researchers attributed to the relatively small number of pediatric exams available for model development.
The findings suggest that non-contrast MRI may contain subtle information about tumor biology that could be used to predict contrast enhancement, including features that may not be visible to human readers. However, the researchers emphasized that the model is not currently accurate enough to replace contrast-enhanced MRI.
“Deciding whether a patient needs a contrast injection is a common judgement in brain tumor imaging, yet it is not always straightforward, or even possible,” Ruffle said. “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.”
The researchers said the long-term goal is to support more individualized decisions about contrast administration rather than eliminate contrast from brain tumor imaging altogether.
Read more here.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.