Play Live Radio
Next Up:
0:00
0:00
Available On Air Stations
Health, Science & Tech

With Big Data And Little Sleep, CMU Students Work To Accelerate Pace Of Brain Mapping

Fig1_ForArticle_JPEG.jpg
Courtesy Carnegie Mellon University
/

Graduate students at Carnegie Mellon University have discovered a method for classifying nerve fibers that could accelerate the pace of brain-mapping projects.

As part of the University’s first ever Neurohackathon, students studying computer science, machine learning and neural computation put their skills to use analyzing large datasets generated by researchers on campus.

Five teams competed in the two-day event, which was organized by faculty involved with CMU’s multi-disciplinary brain research center, BrainHub.

The winning team analyzed data from more than 130,000 three-dimensional MRI images showing the paths of nerve fiber bundles through the brain.

IMG_5485.JPG
Credit Liz Reid / 90.5 WESA
Ph.D. student Mariya Toneva presents her team's research to the judges as part of Carnegie Mellon University's first ever Neurohackathon.

“For example, there is a big one that goes horizontally, connecting the left and right hemispheres,” said Ph.D. student Ying Yang, who is studying machine learning and neural computation.

For the most part, every human brain has the same kinds of bundles that take the same kinds of paths. Currently, the only way to classify individual bundles into different categories is by hand; experts in neuroscience carefully analyze the MRI images in order to label the nerve fiber bundles. The work, though tedious, is important. Mapping the brain is the first step toward developing a deeper understanding of disorders such as Alzheimer’s disease and schizophrenia.

“Our goal would be automate this process of labeling these trajectories,” said MariyaToneva, a Ph.D. student also studying machine learning and neural computation.

Using the average of 130,000 images, the team of six students developed an algorithm that would allow a computer to identify and label 55 different types of nerve fiber bundles.

When they tested their algorithm against data from one individual’s brain, they found it was 85 percent accurate in identifying bundle types.

“We were so happy when we saw it works well for individuals,” said Yang.

Geoff Gordon, associate professor in the Machine Learning Department at CMU, was one of three judges. He said he was impressed the team's rigorous, yet simple approach, especially considering they only had 28 hours to work on the problem.  

IMG_5473_0.JPG
Credit Liz Reid / 90.5 WESA
Recent Ph.D. graduate Prasanna Kumar Muthukumar, Ph.D. student Waleed Ammar and Master's student Volkan Cirik put the finishing touches on their presentation ahead of judging at the 2016 Neurohackathon.

“They did a better job of setting out to answer a problem and laser-focus honing in what is a good solution to it,” Gordon said.

The team even found and corrected errors in the dataset they were working with. Out of 130,000 nerve bundle images, 43 were mistakenly labeled twice and placed into two different categories.

One student from the winning team will receive a free semester’s tuition to continue work on how to automate the labeling of nerve fiber bundles, though they said they still have to figure out who that student will be.