New MRI Algorithm Cuts Scan Time by Two-Thirds

A far quicker MRI scan is on the horizon. A new algorithm developed by MIT’s Research Laboratory of Electronics cuts the imaging time by two-thirds, though they’re still working on the back end processing time. Authors of the research, which is scheduled for publication in the journal Magnetic Resonance in Medicine, say that a 45-minute scan can be done in 15 minutes without compromising much of the quality.

To make the scan faster, the algorithm uses information from the first contrast scan to help it produce subsequent images. The scanner builds on the information it already has (in the form of a basic outline) to produce images from the raw data, and this shortens the scanning time.

To make the outline, the software looks for common features of the particular scan conducted, like anatomical structure. “If the machine is taking a scan of your brain, your head won’t move from one image to the next,” Elfar Adalsteinsson, associate professor of electrical engineering and computer science and health sciences and technology, said in a statement. “So if scan number two already knows where your head is, then it won’t take as long to produce the image as when the data had to be acquired from scratch for the first scan.”

The algorithm predicts the position of tissue boundaries from information gleaned in the first scan, for use in the subsequent scans. “Our idea was that in clinical MRI, what you do is acquire images with different contrast ratings to increase the diagnostic power, and you can use different contrast,” said Berkin Bilgic, first author and a graduate student at MIT. “One [contrast] might be emphasizing grey matter and the other might be emphasizing white matter. This is the same underlying image but it is a different contrast rating on different tissue, so it looks quite similar but the colors are different. So we made use of this statistical correlation of the images to improve the construction.” He said that they used the fact that the tissues share similar edge properties, along with using the popular compressed-sensing algorithm technique.

Their goal, said Bilgic, was a minimal impact on reconstruction quality. He said the result was that the image quality suffers slightly, but it’s hard to notice when viewing it. “If you look at it pixel by pixel, you can see the difference, but visually it is difficult to see. Only when you take the difference and scale it up ten times you can see the difference,” he said.

Bilgic said they compared their technique to the sparse MRI algorithm, and found that theirs increased reconstruction quality up to four times more than the sparse quality.

Another issue is the processing time that turns the raw image data into a final scan. “Right now initial implantation takes quite long,” said Bilgic. That time is currently at four to five hours for three slices, whereas a traditional scan is one to two minutes for the same. “It’s still too long to implement it in real life,” he said.

Since the study finished a year ago, the group has continued to try to improve the data processing speed by using graphics processing units (GPUs) from gaming units instead of standard computer processors.

“A normal computer would just do one operation at a time, and GPU it can do many operations simultaneously. That’s why we think it should help us improve [the reconstruction time],” he said.
The algorithm would work in any scanner, and they’re trying to bring the technology to market. Bilgic said they have an agreement with Siemens, and are talking with them about possible implementation. Siemens officials said they have no official comment on the MRI algorithm or the agreement at this time.

Editor’s note: A previous version of this article incorrectly stated the processing time of a traditional scan as hours, rather than minutes. We regret the error.