First Ever Real-Time MRI Video Captures Images of Body's Interior in Just 20 Milliseconds

By Clay Dillow Posted 08.30.2010 at 12:30 pm 3 Comments

A Real-Time MRI Image of a Heart Frahm / MPI for Biophysical Chemistry

Those long periods of lying completely still inside that intimidating MRI tube may soon be a thing of the past. Employing some tricky math and some heavy-duty computing power, researchers at the Max Planck Institute in Göttingen have developed a new MRI method that renders images in just one-fiftieth of a second, fast enough to capture organs and joints "live" for the first time.

Magnetic resonance imaging (MRI) used to take several minutes to capture a single image, requiring patients to remain perfectly still in order to get a clear cross-sectional image of an internal organ or joint for use in diagnosis. This meant that while MRI was far preferable to X-ray -- which requires the patient to be doused with radioactive waves -- the technique was highly sensitive to any movement by the patient and could not actually capture an internal process in video-like motion.

New techniques such as FLASH (fast low angle shot) have greatly increased the pace at which an MRI machine can capture an image, but none has come close to being able to capture real-time processes like the beating heart or joint movement inside the body live. But this newest technique does exactly that by reducing the amount of time the machine needs to capture individual frames to just 20 milliseconds.

To enable such a leap in image rendering times, the team had to engineer algorithms that allow a computer to assemble a complete image from a less-than-perfect set of image data. Coupled with radial encoding of the spatial information, which makes the images insensitive to movements, some clever mathematics allows the MRI machine to fill in gaps in the image data such that a complete frame can be pieced together from as little as five percent of the data normally needed to create an image.

The new MRI method can be easily adapted to modern imaging equipment, but the process is hindered by computing power; a real-time MRI rendering of a beating heart for one minute will produce 2,000-3,000 images that must each be rendered and stitched together into a seamless video, eating up 2 gigabytes of memory. The whole process would take the program about 30 minutes to complete, so researchers still haven't reached the point of real-time, live MRI video, but they're getting there.

In the meantime, half an hour isn't so long to wait for a moving MRI video that should help researchers more accurately diagnose everything from joint injuries to coronary heart disease.

Previous Article: Use Microsoft Surface to Control a Swarm of Robots With Your Fingertips

Next Article: Taking Cues From Medical Tech, Big Oil Could Use Nanoparticles to Hunt for Leftover Crude in Spent Wells

3 Comments

Link to this comment

rpenri

08/30/10 at 10:38 pm

This is good news. There's been a study done that shows that CT scans and other forms of x-ray scanning have more than doubled for the average patient.

Some people can get scanned over and over again and there's really no precaution or limit on how much a person is exposed to the rays.

A lot of doctors, instead of using old fashioned methods of detecting injuries like physical examinations just opt for the CT scan because they are lazy or they are swamped and aren't able to devote enough time to each patient.

Next thing you know, the patient has been sent to one specialist who orders one scan, gets referred to another specialist who orders another scan, and it goes on and on and on.

Link to this comment

Red_oddity

08/31/10 at 6:41 am

So, can't they just upgrade the computing part using Tesla's in a Cubix GPU Expander?

http://www.nvidia.com/object/medical_imaging.html
http://www.cubixgpu.com/Products/Pro2?PHPSESSID=7d2abf5bd8ff85435193374fa84321a7

Link to this comment

vincent fortune

02/21/13 at 7:35 pm

Well, not quite. The process that takes the time and power is the registration of the images. That is- each MRI slice may be 4mm thick, in the form of a Dicom image, with a gap between the slices. So 512x512x(7) : 1835008 voxels per slice, as an example.Multiply this by say, an average of 25 slices per scan and you get to 45875200 voxels, each with 1024 grayscale levels, each occupying a x,y and z position,which is re-mapped every 20 m/s. The whole process would be simultaneously repeated in all three planes - axial, sagittal and coronal to make a coherent model without stepping between the slices.Think of it like a massive spreadsheet. That's a lot of Giga/Petaflops.

You are referring to surface/texture rendering. Using Tesla's or SS10000's speeds up the display, rotational screen movement and texturing of 3D models,but it comes down to raw computing power in the form of a Beowulf cluster or similar with infiniband connections to distribute the processing load over many thousands of cores. Its just that there are two different jobs to do, and perhaps the term "rendering" used in the article is not the best one to use. Post-processing would have been more accurate. Hope this explains things, I've been working on a parallel project for the last four years.

Popular Tags

Science

All Tags All Photos All Videos

Technology

+ More Photo Galleries

June 2013: American Energy Independence

Five amazing, clean technologies that will set us free, in this month's energy-focused issue. Also: how to build a better bomb detector, the robotic toys that are raising your children, a human catapult, the world's smallest arcade, and much more.