New supercomputer 'like magic' during world-wide MRI competition



UAB’s new supercomputer “worked like magic” as University professor Dr. Hassan Fathallah-Shaykh participated in a world-wide competition to evaluate MRI images of the brain.

Fathallah-Shaykh, a professor in UAB’s Departments of Neurology, Mathematics, Integrative, Developmental and Cell Biology, Biomedical, and Electrical Engineering won the 2016 BRATS Competition in Athens, Greece.

BRATS is short for Multimodal Brain Tumor Image Segmentation Challenge. Dr. Fathallah-Shaykh’s team, along with fellow Blazer biomedical Engineering graduate student Fabio Raman, tied with two other groups to top the field of 19. The use of UAB’s new supercomputer helped to separate Dr. Fathallah-Shaykh from the rest of the field, he said.

Last year, UAB IT installed the fastest supercomputer in the state — and one of the fastest in the Southeast. UAB increased computing speed for researchers from 10 teraflops to 110 teraflops, and increased storage from 0.7 petabytes to nearly 7 petabytes.

“The supercomputer worked like magic, without any glitches. It is evident that I would not have been able to compete without the supercomputer resource,” he said.

The BRATS Challenge consisted of the MRIs of 191 brain tumor patients diagnosed with low and high grade gliomas in different parts of the brain before and after surgery. Teams were tasked with determining the location of the segments of each MRI image that include enhancing tumor, FLAIR signal changes, necrosis, and tumor core, if any.

Dr. Fathallah-Shaykh’s method was the only method that was interactive, semi-automated and did not use statistical learning or neural networks.  It is designed such that the user can easily check the results and easily make modifications, if needed.

The method consists of two parts, and Fathallah-Shaykh used UAB’s research computer Cheaha to run first part of this process, which was the segmentation with NMF-LSM (Non-Negative Matrix Factorization and modified Level Set Method). He was granted 265 processors; all calculations were competed in about 12 hours — a process that without the supercomputer would have taken days to perform.

The performance of Dr. Fathallah-Shaykh and his team with the use of the supercomputer is a further sign that UAB’s “investment in research computing is paying off. UAB IT is committed to giving our researchers and faculty world-class technology, so that they can innovate and change the world," said Curt Carver, UAB’s VP of Information Technology.