AI-Based Scans Could Eliminate the Need for Using Gadolinium in MRI

According to a report recently presented at the Radiological Society of North America’s annual meeting, important clinical data demonstrated by low-dose gadolinium MRI exams can now be effectively conveyed through algorithm-enhanced MR images, possibly providing a safer alternative to using contrast elements when performing MRI.

The use of gadolinium-based contrast agents in imaging practices has recently come under federal scrutiny, and according to the Food and Drug Administration, it can remain in the brain for months or years after being administered. The National Institutes of Health have also warned that patients who have suffered from acute stroke could be prone to gadolinium leaks upon undergoing imaging exams.

Researchers at Stanford University are using deep learning to produce images that are equally effective and insightful as gadolinium-based ones. Through the use of convolutional neural networks, the researchers have created an AI-based algorithm by programming images from 200 patients that had undergone contrast-enhanced MRI exams for a range of indications. Each patient had a pre-contrast scan, which contained no gadolinium; low-dose scan, an image that was produced after 10 percent of the standard gadolinium dose had been given; and a full-dose scan, which contained 100 percent dose administration.

The algorithm proved to successfully distinguish full-dose scans from zero and low-dose scans. Neurologists then weighed in on the contrast enhancement quality. They found that overall scan quality across the low-dose, full-dose, and algorithm-enhanced MR images were “not significantly different.” These results