Researchers Sniff at Computer-Aided Nasal Cavity Surgeries
Scientists team with doctors to simulate human respiration on supercomputers
Principal Investigator: Andreas Lintermann, Fluid Mechanics and Institute of Aerodynamics, RWTH Aachen University (Germany)
HPC Platform: Hermit of HLRS
Medical professionals want supercomputing research to answer questions related to one of humanity’s most basic needs — breathing. Luckily, Andreas Lintermann and a group of researchers at RWTH Aachen University are employing computing resources at the High-Performance Computing Center Stuttgart (HLRS) to do just that.
Scientists and medical professionals know about the human respiratory system’s basic functions and what organs and tissues are involved. More mysterious, though, is how physical processes relate to respiration, which presents new challenges to doctors in diagnosing many breathing problems affecting their patients.
“We can have a lower risk for surgeries with a higher success rate by eliminating trial and error,” Lintermann says. By using supercomputing simulations, researchers try and create accurate simulations of individual nasal cavities, ultimately working toward surgeons using these simulations at the commercial level before ever touching a patient at the operating table. Medical professionals try to avoid surgery on nasal cavities at all costs, and will only perform such operations if patients are suffering from severely limited breathing abilities. Such issues, like septum deviations, sinusitis and extreme cases of allergies, require surgeons to perform several different operations.
Electrosurgery, where a needle is inserted into swollen nasal tissue and a high frequency current is used to help with swelling, or cold ablation, where tissues are removed by extremely low-temperature plasma, are just a couple of the procedures that could benefit from simulations before a patient goes to the operating table.
Though simulations like Lintermann’s have very practical applications in the medical industry, the intensive computational calculations currently prevent commercial use. The complexity of organ and tissue functions, combined with the need for detailed interdisciplinary collaboration, pushes researchers to turn to the world’s fastest computers to create these simulated operating rooms. Lintermann and his collaborators were granted time on the Hermit supercomputer at HLRS. Hermit is a Cray XE6 capable of more than 1 petaflop, or 1 trillion calculations per second.
Lintermann and his team collaborate with medical professionals from the University Hospital Aachen (UKA), rhinology experts from the Institute of Statistics and Medical Informatics Cologne (IMSIE), the virtual reality group from the Computing Center of the RWTH Aachen University (VR RZ RWTH), and computer scientists at HLRS.
Inhaling data, exhaling results
Researchers must start with real-world data to create nasal cavity simulations. Computational scientists receive computer tomography (CT) scans from collaborators at hospitals. These CT scans contain information about tissue, bone and air density, which define a cavity’s form. Researchers first pre-process the image in Medical Interaction Toolkit (MITK) simulation environment, developed at the German Cancer Research Center (DKFZ) in Heidelberg. Lintermann extended the toolkit so it is capable of extracting surfaces from CT data of nasal cavities by breaking the data into computerized segments. Subsequently, the segment defining the volume of the nasal cavity is contoured to obtain a surface. To solve the equations of fluid motion on computers, the space needs to be broken into small, individual cubes. Based on the obtained surface, the cubes are split into smaller and smaller units until a desired resolution is reached. The high memory demand for these simulations necessitates the use of supercomputing resources like Hermit to efficiently create such grids for the simulation. Most of these grids contain billions of small cubes, each one containing individual information about the fluid flow and temperature within the nasal cavity.
Once researchers have set up the grid for the cavity itself, they use their self-developed Zonal Flow Solver (ZFS) to put their simulation in action. When ZFS is used on Hermit, it allows for billions of cells for every simulation, where the grids are created in only seconds. Lintermann notes that without the power of parallel computing, simulations like this would be impossible to do in a timely manner. With parallel computing, Lintermann and his collaborators can assign processors to small subsets of cells. Each processor simulates a specific part of the flow in the nasal cavity as the simulation runs, and communicates with the others during the solution process.
For nasal cavity simulations to be applied to real-world surgery environments, researchers must also take temperature and air pressure into account, doubling the computational demand for each simulation. Lintermann also notes that taking fine dust particles or diesel engine emissions into account could also greatly improve diagnoses, but doing these all-encompassing simulations in a timely manner requires more code optimization and computing power.
Breathing easier in the future
The continued support of HLRS and CRAY helps Lintermann and his group to push their algorithms forward, which drastically helps to make the most of Hermit’s computational power. The team works to not only improve the speed of ZFS on Hermit, but works to speed up communication between processors. The concept of Moore’s Law, which states that computational power doubles roughly every 18 months, encourages Lintermann that through code optimization and supercomputing power advancement, these virtual surgery environments will be commercially available sooner rather than later.
“We have a surgeon performing a surgery with pre and post-CT images and then compare the results to see if we really got an increase in breathing capability,” Lintermann says, as he explains one way in which his research could be applied in hospitals. “In this way we can evaluate current standardized surgery methods.”
Another method would be for surgeons to identify specific regions of CT-scans they would like to focus on, and computational scientists using shape optimization on those areas. Surgeons could then get a quick turn-around time, and be able to use computational results to cross-check diagnoses and help guide their incisions during surgery.
Though Lintermann’s simulations already scale well on current machines, the team is always looking for faster computers to create more comprehensive simulations. The research team is currently working on more complex simulations that also include aeroelastic—the interplay of the elasticity of tissue with air flows—and material properties of nasal tissues. By simulating material movement, researchers can use these simulations for diagnosing and treating common problems such as sleep apnea and snoring.
—by Eric Gedenk