The MRI machines and CAT scans, blood analyses and gene sequencing tools that are used to help diagnose our illnesses rely on advanced computing to extract knowledge out of molecular markers and reflected laser beams.

“Medical advances may seem like wizardry,” remarked Harold Varmus, former director of the National Institutes of Health (NIH). “But pull back the curtain, and sitting at the lever is a high-energy physicist, a combinatorial chemist, or an engineer.”

Ching-Long Lin, professor of mechanical and industrial engineering at the University of Iowa and a research engineer at IIHR-Hydroscience & Engineering, is part of a multidisciplinary team working to develop a new tool to image, understand, and diagnose how air flows through the thousands of branching passageways of the lung, and how abnormalities can lead to illness.

“Our approach to understanding the airflow and particle transport in the human lungs is quite novel,” Lin said. “We use computed tomography [CT] images to construct realistic human lung models, and then we use computational fluid dynamics models to simulate the airflow through the lung.”

His computer simulations on Lonestar and Ranger at the Texas Advanced Computing Center (TACC) combine 20 years of experience modeling turbulence and computational fluid dynamics (CFD) with cutting-edge medical imaging technologies to create a framework that will help doctors understand what causes asthma, how exposure to environmental pollutants alter the development of children’s lungs, and how the addition of helium to aerosol drugs can make pharmaceuticals more effective.

HPC and Physiological Modeling

The growth in the processing power of high-performance computing (HPC) has led to dramatic improvements in the ability to create three-dimensional models of the human body. As recently as 10 years ago, it was only possible to simulate the trachea and a single bifurcation, or generation, of the lung. Today, Lin’s algorithms on Lonestar and Ranger simulate up to 23 generations of branching airways to create a mesh model of tremendous complexity and usefulness. [See below for the evolution of lung simulations]

The largest of these lung branches are centimeters wide; the smallest are measured in millimeters. Since airways at different scales have important, interrelated features, all of them need to be integrated into a multi-scale whole — a feat that had never before been accomplished because of the inherent algorithmic challenges and extreme computational demands.

Lin’s contribution to the project, and to multi-scale modeling overall, is a numerical scheme coupling very accurate 3-D models of the lungs (generated by the CT scan) with less accurate but more widespread 1-D models (derived from computer simulations), combining the best of both worlds and representing how they function together.

The method lets the team determine the boundary conditions at the outlets of the smallest branches with far greater accuracy. This allows researchers to model how air flows through 23 generations of branches, where it becomes turbulent, and where particles are deposited.

This type of multi-scale modeling is invaluable not only for simulating pulmonary airflow, but also for other physiological systems, including . The data derived from Lin’s framework, because it is based on actual high-fidelity CT scans, is subject-specific, describing the lungs of a particular individual. This is critical for treatment purposes, and helps doctors explore diseases like asthma by comparing the airways of an asthmatic patient to a healthy subject.

“The multi-scale, CFD framework is essential for delivering this technology 'to the bedside' because it allows us to select a combination of the 3-D and 1-D domains to give fine scale computation (3-D) in critical areas, and coarse scale computation (1-D) in the remainder of the lung,” said Merryn Tawhai, senior research fellow at the Auckland Bioengineering Institute and Lin’s collaborator on the project. “This minimizes model size, reduces run time, and makes the process of patient-specific modeling feasible.”

For instance, drug-makers have long believed that mixing helium with drug aerosols can increase the effectiveness of certain pharmaceuticals, but they weren’t sure of the cause.

“The pharmaceutical industry believes that by mixing drug aerosols with helium, particle deposition can be enhanced, and now we actually have some idea why,” Lin said. “Because helium is a lighter gas than air, if we inhale helium, the interface between the two gases becomes stable. The particles tend to stay away from the airway walls and can go deeper into the lung. If we can understand the mechanisms behind the deposition, then we can further improve them.”

Another useful insight has emerged from Lin’s airflow simulations of patients with abnormal tracheal structures. In one of their published papers (“Characteristics of the turbulent laryngeal jet and its effect on airflow in the human intra-thoracic airways”, Respiratory Physiology & Neurobiology, 157 (2007) 295-309), Lin showed that some subjects with atypical tracheal geometries had very high shear stress within the lung, which was induced by airflow. “A doctor looked at our result and immediately linked the shear stress to stenosis in the trachea,” Lin recalled. “This is a prediction from CFD, and the result inspired a doctor to look at a real patient to explore whether the correlation exists. So, we are helping each other to understand lung disease.”

Putting Theoretical Knowledge Into Practice

For Lin, this applied, interdisciplinary project is miles away from the fundamental turbulence studies that, until four years ago, had been the focus of his career.

“This is one of the most exciting projects that I’ve ever worked on,” he said. “We have expertise from CFD, engineering, bioengineering, lung modeling, and lung imaging, and we brainstorm and work together to figure out how to integrate it all.”

As with his previous research, these projects require the creative application of supercomputers to model complicated physiological structures with enough resolution to be useful and meaningful. “Without access to Lonestar and Ranger, I don’t think we could accomplish our objectives,” Lin said.

But the release of their novel framework isn’t the end of the road for Lin and his team. Each new discovery opens up as many questions as it answers, Lin said. “Whenever we feel excited about identifying a certain phenomena, we also identify other issues that we haven’t thought about before and we keep improving.”

Still on the horizon, Lin will add a solid mechanical model to account for the nature of the actual lung tissue, and a cell model that integrates pulmonary functions at an even finer level.

With a talented team approaching these problems from many distinct perspectives, and designing ever-better mechanisms for imaging, modeling and understanding pulmonary function, it won’t be long before the system Lin and his colleagues have created is a standard diagnostic tool in all hospitals.

“We have clear hypotheses and we’re always brainstorming to come up with better ideas,” Lin concluded. “I think that’s very important to advance medicine and computational technology.”

******************************************************************************************************************************

This work is supported by NIH BRP HL064368 (PI: Eric Hoffman) and NIH NIBIB EB005823 (PI: C.-L. Lin) under the .

This work has, in part, evolved from the NIH BRP seeking to establish an image-based normative atlas of the human lung, serving as the baseline for the use of imaging in drug and device discovery, as well as outcomes and safety evaluation of resultant new approaches to the treatment of disease.

Aaron Dubrow
Texas Advanced Computing Center
Science and Technology Writer