Machine Learning Used to Automatically Identify Brain Tumors

Summary: Researchers from UT Austin utilize deep learning and supoercomputing to help identify brain tumors.

Primary brain tumors encompass a wide range of tumors depending on the cell type, the aggressiveness, and stage of tumor. Quickly and accurately characterizing the tumor is a critical aspect of treatment planning. It is a task currently reserved for trained radiologists, but in the future, computing, and in particular high-performance computing, will play a supportive role.

George Biros, professor of mechanical engineering and leader of the ICES Parallel Algorithms for Data Analysis and Simulation Group at The University of Texas at Austin, has worked for nearly a decade to create accurate and efficient computing algorithms that can characterize gliomas, the most common and aggressive type of primary brain tumor.

At the 20th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2017), Biros and collaborators from the University of Pennsylvania (led by Professor Christos Davatzikos), University of Houston (led by Professor Andreas Mang) and University of Stuttgart (led by Professor Miriam Mehl), presented results of a new, fully automatic method that combines biophysical models of tumor growth with machine learning algorithms for the analysis of Magnetic Resonance (MR) imaging data of glioma patients. All the components of the new method were enabled by supercomputers at the Texas Advanced Computing Center (TACC).

Biros’ team tested their new method in the Multimodal Brain Tumor Segmentation Challenge 2017 (BRaTS’17), an annual competition where research groups from around the world present methods and results for computer-aided identification and classification of brain tumors, as well as different types of cancerous regions, using pre-operative MR scans.

Their system scored in the top 25 percent in the challenge and were near the top for whole tumor segmentation.

“The competition is related to the characterization of abnormal tissue on patients who suffer from glioma tumors, the most prevalent form of primary brain tumor,” Biros said. “Our goal is to take an image and delineate it automatically and identify different types of abnormal tissue – edema, enhancing tumor (areas with very aggressive tumors), and necrotic tissue. It’s similar to taking a picture of one’s family and doing facial recognition to identify each member, but here you do tissue recognition, and all this has to be done automatically.”

For the challenge, Biros and his team of more than a dozen students and researchers, were provided in advance with 300 sets of brain images on which all teams calibrated their methods (what is called “training” in machine learning parlance).

In the final part of the challenge, groups were given data from 140 patients and had to identify the location of tumors and segment them into different tissue types over the course of just two days.

“In that 48-hour window, we needed all the processing power we could get,” Biros explained.

The image processing, analysis and prediction pipeline that Biros and his team used has two main steps: a supervised machine learning step where the computer creates a probability map for the target classes (“whole tumor,” “edema,” “tumor core”); and a second step where they combine these probabilities with a biophysical model that represents how tumors grow in mathematical terms, which imposes limits on the analyses and helps find correlations.

TACC computing resources enabled Biros’ team to use large-scale nearest neighbor classifiers (a machine learning method). For every voxel, or three-dimensional pixel, in a MR brain image, the system tries to find all the similar voxels in the brains it has already seen to determine if the area represents a tumor or a non-tumor.