Algorithmic Health

Tuesday, August 23, 2011

 

Accoring to the World Health Organization estimates, 17 million people around the world die of cardiovascular diseases each hear. About 9 million of these are women.

Everyone would like to see those numbers drop, of course. George Biros is doing something to make it happen.

Strokes, heart attacks, pulmonary embolisms and other conditions can be caused by thrombosis: blood clots that form in veins or arteries and can travel—sometimes with fatal results—to the brain, heart or lungs. Biros, associate professor in the School of Computational Science and Engineering (CSE), uses high performance computing to study several aspects of the cardiovascular system, from flow in small capillaries to the mechanical properties of ischemic myocardium, in hopes of finding a way to counter thrombosis and its deadly effects.

Biros, who has a joint appointment in the College of Engineering’s Wallace H. Coulter Department of Biomedical Engineering, arrived at Georgia Tech in fall 2008. A native of Greece, he earned his master’s and doctoral degrees from Carnegie Mellon University and completed postdoctoral research in applied mathematics and computer science at New York University. Before coming to Atlanta, he taught at the University of Pennsylvania as an assistant professor of mechanical engineering and applied mechanics, bioengineering, and computer and information science.

Besides research on parallel algorithms for problems in mathematical physics, Biros works with collaborators from Penn in brain image and cardiac image analysis; with collaborators from the University Houston and University of Texas at Austin in fast algorithms for intravascular ultrasound image analysis; and with collaborators from New York University in understanding the interactions for red blood cells with platelets and the onset of blood clotting.

"We want to answer questions about the causes of thrombosis,” Biros says of his current research, “because that is important if you want to design new drugs for thrombosis, or if you want to design stents for bypass surgery or new mechanical heart valves that won’t cause clots to form.”

Experimentation on living patients is nearly impossible, so researchers like Biros have turned to in vitro experiments and computer modeling and simulation for answers. When one uses computer simulations to study different aspects of thrombosis, one basic problem is understanding the hydrodynamic interactions between red blood cells and platelets. But the number of blood cells that affect blood flow and clotting is so enormous that ordinary computers cannot perform the necessary calculations with accuracy and speed. That’s where supercomputing comes in."Around three years ago, we developed these algorithms that allowed us to speed up calculations by five orders of magnitude and allowed simulations of hundreds of thousands of cells—calculations that previously had taken one year to complete could be done in one day or less,” Biros says. “This tool has enabled new discoveries, and I consider it to be the part of my work that has had the biggest impact.”

Blood flow dynamics is only one aspect of that work. Biros is also working on technology to help physicians and radiologists make the most of the information they gather using medical imaging techniques such as CT, MRI and ultrasound. There is a lot of data available that is not used when a radiologist interprets medical images, Biros says, but it’s very difficult to automate the process. The benefits of such technology would go well beyond the individual patient.

"We are using medical images—all this information that we get from clinical datasets—to try to understand better the cardiac physiology, how electric waves propagate in the heart, how the muscle contracts, what are the mechanisms in which the heart  can malfunction or fail,” he says. “Large-scale simulations on supercomputing platforms play an important role.”

Biros says it’s quite hard to design and implement algorithms that simulate complicated physical phenomena. Verification and validation of such software is as difficult as the development of a sophisticated instrument for scientific research.  With the advent of heterogeneous computing platforms—that support streaming, multithreading and distributed-memory computing—there also has come additional challenges in developing languages and compilers that enable high utilization of these novel platforms, while also allowing computer scientists to develop, maintain and debug the programs.

All of Biros’ research involves long-term goals, he admits, not just-around-the-corner leaps. But he hopes that, within the next five to 10 years, he and his fellow researches will be creating tools that could be used routinely—and effectively—in clinical data analysis.