Deep Learning Algorithm Could Reduce Gadolinium Dose for Contrast-Enhanced MRI
Researchers find that a deep learning algorithm can approximate full-dose gadolinium MRI images from low-dose scans
Gadolinium-based contrast agents for MRI are widely used in both research and clinical settings, but questions remain about whether traces of the rare earth left behind in the patient are harmful. Studies have found gadolinium deposits in the brain tissue of patients after intravenous administration of these agents, whereas those never exposed to gadolinium have no such deposits.
It isn't yet known whether the metal is detrimental to the brain, and no such evidence has been found, but many radiologists want to mitigate potential patient risks by reducing gadolinium dose if possible. In a new study, a team of researchers from Stanford University tested a deep learning algorithm that can approximate full-dose gadolinium MRI images from low-dose scans. The results were presented Nov. 26 at the annual meeting of the Radiological Society of North America in Chicago.
The study included 200 patients who received clinically routine contrast-enhanced MRI exams for a variety of medical reasons. The authors performed three MRI sequences: zero-dose (precontrast), low-dose (10 percent dosage administration), and full-dose (100 percent dosage administration). They trained a deep convolutional neural network to learn how to approximate full-dose images using the zero-dose and low-dose images.
The image quality of the low-dose images that were enhanced by the algorithm to approximate full-dose images did not differ significantly from the images taken using the full dose of contrast agent. These initial results suggest the possibility of drastically reducing gadolinium dose without sacrificing the clinical value of contrast-enhanced MRI exams, although more research must be conducted. The authors plan to evaluate the algorithm further in a clinical setting, across a broader range of MRI scanners and contrast agents.