An MRI scan of a patient’s brain is currently one of the best sources of information for a neurologist investigating a tumor. The scan shows the location of a tumor, the composition of the tissue, and how it may be entangled with other tissues in the brain, from blood vessels to nerve bundles. It’s the kind of information that is key for the treatment planning.
However, for all it is worth, the scan does not show all the abnormalities. It’s up to medical professionals to predict the nature of the tumor—for example, the depth it penetrates into brain tissue or even the tumor grade. Sometimes there is disagreement among doctors about these assessments. Locked up inside a patient’s skull, a tumor is a difficult subject to get to know from MRI images alone.
As a Ph.D. student ICES, Amir Gholaminejad focused his thesis research on developing a suite of algorithms that work together to make predictions about tumor behavior from MRI images. Using tumor biophysics, mathematics and computer science, his work is helping doctors extract more information from medical images.
“Starting from scratch we asked how we could incorporate our current understanding of tumor bio-physics with medical image analysis,” said Gholaminejad, who is now continuing his work as a postdoctoral researcher at the University of California, Berkeley. “ This requires expertise in different areas of computational science and combining them all together into one piece is what interested me.”
Gholaminejad's work was honored with both the ICES Outstanding Dissertation Award and UT's 2018 Outstanding Dissertation Award in Mathematics, Engineering, Physical Sciences, and Biological and Life Sciences, recognizing his innovative approach to biophysics-based medical image analysis and the promise it has as a potential medical tool.
In a nomination letter to the university’s dissertation award committee, Todd Arbogast, a professor of mathematics and the chair of the ICES graduate studies committee, highlighted how Gholaminejad’s thesis research contributed to a diverse array of scientific advances while focusing on a larger goal of tumor modeling.
“Our program chose this dissertation for nomination because it describes major, outstanding advances in a wide range of fields, including imaging science, inverse problems, biophysics modeling, numerical analysis, scientific computing, and high performance computing,” Arbogast wrote. “We believe that Dr. Amir Gholaminejad’s dissertation is outstanding in every way, and that it is a credit to The University of Texas at Austin.”
The Outstanding Dissertation Award is the latest award Gholaminejad’s thesis research has received. A publication based on the research won the best student paper award at SC 2017, the International Conference for High Performance Computing, Networking, Storage and Analysis. In 2015, a poster presenting the research in its earlier stages earned the Association for Computing Machinery Student Research Competition Gold Medal.
The developed framework relies mainly on four algorithms that drive the tumor modeling and image analysis process. The first algorithm analyzes the MRI image to distinguish between different types of brain anatomies; the second applies inverse modeling to predict how the tumor grew from its initial state; the third couples the first two algorithms together; and the fourth optimizes the whole process so that it can be swiftly carried out on a high performance computers, with the processing speed taking only a few minutes on supercomputers.
At Berkeley, Gholaminejad is continuing the research by incorporating artificial intelligence into the process. An important input to the framework is prior knowledge about specific tumor parameters. By adding AI technology, Gholaminejad hopes that the framework could have better prior information given similar instances on other patients.
Another aspect of the future work is uncertainty quantification. “all models are wrong to some degree” emphasizes Gholaminejad. Therefore it is important to quantify the uncertainties in the prediction by taking into account measurement and model errors. With the skills he learned at ICES, he said he feels well equipped to keep developing his program from a thesis research project to clinical tool.
“I think that getting a Ph.D. from ICES prepares you to tackle any problem in computational science,” Gholaminejad said.