Jun 19, 2018 | By Thomas

Researchers from the Massachusetts Institute of Technology (MIT) in Cambridge have created a machine-learning algorithm called "VoxelMorph” that they say makes the process of medical image registration more than 1,000 times faster.

MIT researchers describe a machine-learning algorithm that can register brain scans and other 3D images more than 1,000 times more quickly using novel learning techniques. (Source: MIT Media Lab)

Medical image registration is used by doctors daily to see the differences in two MRI scans. It is a valuable assistant for the medical experts to compare and analyze anatomical differences in two scans over time. This process, however, can often take up to two hours or more, as traditional systems have to align the scans pixel by pixel. Now MIT researchers have created a machine-learning algorithm that can register brain scans and other 3D images more than 1,000 times more quickly using novel learning techniques.

Their algorithm, called VoxelMorph, learns while registering thousands of pairs of images to create the perfect alignment for each scan. In doing so, it acquires information about how to align images and estimates some optimal alignment parameters. After training, the algorithm uses the parameters it learned to “map all pixels of one image to another, all at once.” This brings medical image registration time to a minute or two using a normal computer, or less than a second using a GPU.

The team’s research is part of two papers scheduled to be presented at the Conference on Computer Vision and Pattern Recognition (CVPR) June 18-22 in Salt Lake City, Utah, and the Medical Image Computing and Computer Assisted Interventions Conference (MICCAI) September 16-20 in Granada, Spain.

“The tasks of aligning a brain MRI shouldn’t be that different when you’re aligning one pair of brain MRIs or another,” says co-author on both papers Guha Balakrishnan, a graduate student in MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) and Department of Engineering and Computer Science (EECS). “There is information you should be able to carry over in how you do the alignment. If you’re able to learn something from previous image registration, you can do a new task much faster and with the same accuracy.”

MRI scans are basically hundreds of stacked 2D images that are gathered together to form a massive single 3D images. Therefore, it’s very time-consuming to align all voxels or pixels in the first volume with those in the second. Current algorithms also don’t learn from each scan. After each registration, they dismiss all data pertaining to voxel location. “Essentially, they start from scratch given a new pair of images,” Balakrishnan says. “After 100 registrations, you should have learned something from the alignment. That’s what we leverage.”

VoxelMorph is powered by a convolutional neural network (CNN), which is used all the time for image processing. In order to train VoxelMorph, the algorithm was fed 7,000 publicly available MRI brain scans to learn from. After that, the team fed the system 250 additional scans to test it. During training, brain scans were fed into the algorithm in pairs. Using a CNN and modified computation layer called a spatial transformer, the method captures similarities of voxels in one MRI scan with voxels in the other scan. Through the learning scans, the algorithm learned about groups of voxels which it uses to calculate optimized parameters that can be applied to any scan pair.

When VoxelMorph is fed a new MRI brain scan, the system uses a mathematical “function” to rapidly calculate the perfect alignment of every voxel in both scans. This system only needed one evaluation to process the images. The researchers found their algorithm could accurately register all 250 new brain scans within two minutes using a traditional central processing unit, and in under one second using a graphics processing unit.

Importantly, VoxelMorph is an “unsupervised” algorithm meaning it doesn’t require additional information beyond image data. It also guarantees the registration “smoothness,” as it doesn’t produce folds, holes, or general distortions in the composite image. Across 17 brain regions, the refined VoxelMorph algorithm was proved to be just as accurate as a commonly used state-of-the-art registration algorithm in a fraction of the time.

The researchers said their findings could allow image registration to take place during operations, and surgeons could potentially register scans in near real-time, getting a much clearer picture on their progress.

“Today, they can’t really overlap the images during surgery, because it will take two hours, and the surgery is ongoing,” Adrian Dalca, a postdoctoral research student from Massachusetts General Hospital in Boston and a co-author on both papers, said in the same MIT news report. “However, if it only takes a second, you can imagine that it could be feasible.”



Posted in 3D Scanning



Maybe you also like:


Jordi wrote at 6/20/2018 7:10:15 PM:

Where's the relation of this very interesting news to the 3D Printing world ?

Leave a comment:

Your Name:


Subscribe us to

3ders.org Feeds 3ders.org twitter 3ders.org facebook   

About 3Ders.org

3Ders.org provides the latest news about 3D printing technology and 3D printers. We are now six years old and have around 1.5 million unique visitors per month.

News Archive