Using Math to Map Lung Function

Tuesday, November 12, 2013

The following feature is highlighted in the 2012-2013 University of Iowa Graduate College Annual Report.

 
 
 
 
 
 
 
 
 
 
 
 
If diagnosed with lung disease, you might use an inhaler to breathe medicine into your airways. The goal is to deliver the right amount of medicine to the right places in your lungs. However, the medicine's route and endpoint in the lungs are unclear, increasing the guesswork in respiratory therapy. Lack of information about what's happening in airways can also hamper diagnosis.

Digital lung model

University of Iowa Professor Ching-Long Lin and his collaborators have been developing a digital model of the human lung PDF icon to simulate particle movement through the lungs. The model can be used to track any kind of particle—pollutants or pharmaceuticals. Using the model, the team studies lung air flow to aid diagnosis and to learn more about effective delivery of medicine to diseased passageways.

“Air needs to be uniformly distributed throughout the lungs, but asthmatic people, for instance, have more air transported to the upper lobes,” Lin says. “We want to predict a medical condition before it happens, so we try to identify biomarkers that allow us to distinguish between normal lungs and diseased lungs.”

Lin used computed tomography (CT) scans acquired by Eric Hoffman, UI professor of radiology, to construct a digital model of the human lung. Next, Lin utilized his background in computational fluid dynamics (CFD) to simulate air flow and aerosol transport through the lungs.

Modeling lung air flow presents a particular computational challenge because of the complexity of airway structure, the large amount of data, and long computational time. To compile and anaylize such massive amounts of information, Lin's team used the UI’s high-performance computing cluster—dubbed Helium—alongside national computing resources from the Extreme Science and Engineering Discovery Environment.

The resulting digital lung model consists of several building blocks, including geometric modeling, CFD, particle tracking, and epithelial cell modeling. The model integrates the data from these sources using a process called image registration, which enables researchers to merge data obtained from different measurements of the same subject. Additional processing that transforms the data into a visual format—a technique called visualization—enhances researchers' understanding of the data.

Better, faster visuals

Nathan Ellingwood, a Ph.D. candidate in the Interdisciplinary Graduate Program in Applied Mathematics and Computational Sciences PDF icon, is researching ways to improve the model's particle tracking, image registration, and visualization. His primary tool is a computer video card called a GPU (Graphics Processing Unit).

GPUs, commonly used in computer games, perform the same set of instructions on different data sets of a single subject. For example, imagine several series of photos of a man running, each series taken at the same time, but from different angles. A GPU could analyze each photo series and then merge them into one information-rich photo series of the man running. This is called parallel computing. A major advantage of using video cards for parallel computing is their relative low cost, ranging anywhere from $100 to $2,000.

“(Image registration) is an important tool for medical imaging analysis. It matches two images of the lungs to derive regional ventilation that affect aerosol distribution in the lungs,” Ellingwood says. “Typically, it takes about two and a half hours to register a pair of lung images using a high-end 32-core CPU. With the aid of a GPU, it now takes only 50 minutes. This is critical when analyzing a large amount of image data.”

Using a GPU for fast computation opens additional opportunities for researchers, such as real-time visualization of results. “In particle tracking, we’re releasing tens of thousands of particles and we’re tracking them individually,” says Ellingwood, a Presidential Graduate Research Fellow. “My code for particle tracking has really made a big difference, running 24 times faster than the CPU-based version.”

With these real-time GPU results, the next step is to create a visual representation of particle flow in the digital lung model. To do this, Ellingwood and his colleagues use a technique called augmented reality. In augmented reality, a camera takes a picture in real time, and the researchers superimpose a graphic on the picture. The augmented reality technique allows the researchers to rotate the images, helping them learn more about what's happening in the virtual airflow field.

This digital lung model remains in the developmental stages and is not yet available for treating patients. “We have a long term goal of making our digital lung model commercially available,” Lin says. “Right now, we don’t have a nice user interface. There is a learning curve for our students to use it right now.”

International perspective and collaboration

Ellingwood received a National Science Foundation (NSF) East Asia and Pacific Summer Institutes for U.S. Graduate Students grant in 2011 to study abroad and a Graduate College T. Anne Cleary International Dissertation Research Fellowship in 2013.

Ellingwood used the grants to initiate collaborations and enrich his research experience in an international setting. He worked with Matthew Smith, an internationally known expert in GPU computing who is a professor at National Cheng Kung University in Taiwan.