At Liberty University in Lynchburg, Virginia, a mechanical engineering team received a Microsoft AI for Health COVID-19 grant that provided access to Microsoft Azure Cloud, Microsoft High-Performance Computing (HPC) capabilities, and Microsoft AI for Health data science experts. The grant helped them study how a new respiratory therapy device can help COVID-19 patients and their caregivers.
Research projects at the Liberty University School of Engineering went online during the COVID-19 health crisis. Associate professor of mechanical engineering Wayne Strasser, P.E., Ph.D., and his lead research assistant, Ph.D. student Reid Prichard, were working at the cutting edge of computational fluid dynamics (CFD) when their team moved its research to the cloud. They were studying the effectiveness of Vapotherm’s FELIX-1, a new respiratory therapy device that can be used with Vapotherm’s existing High Velocity Therapy (HVT) device to help keep COVID-19 patients off mechanical ventilators while also protecting caregivers.
In April 2020, the team received a Microsoft AI for Health COVID-19 grant, created to empower nonprofits, researchers, and organizations that are tackling some of the toughest challenges in global health. The grant provided them with Microsoft Azure Cloud and High-Performance Computing (HPC) capabilities and access to Microsoft AI for Health data science experts.
“It was a confluence of the urgent need to run COVID-19 studies and the Microsoft grant availability,” says Strasser. “This put us in the unique position to become the only group in the world studying this specific device. We wanted to drive this research knowing it will have ramifications not just for COVID-19 patients, but with other airborne contagions in the future.”
Strasser and Prichard were trying to solve a common problem related to HVT, which delivers supplemental oxygen, heated and humidified, to patients in respiratory distress. Traditionally, patients wear a disposable surgical mask over the HVT nasal cannula device to protect others in the room from their exhalations. But not only do patients often find the masks hot and itchy, the masks also frequently leak around the edges.
The FELIX-1, however, is a plastic face mask that patients can wear over an HVT nasal cannula in place of a disposable mask. It uses suction to clear the air around a patient’s mouth and nose. Strasser and Prichard used CFD to understand how effective the device is at capturing particles exhaled from the patient’s nose and throat and where any escaped particles move within a hospital environment.
Their research consisted of simulating one hospital room, two patients, and four caregivers. A critical step in the CFD model was establishing lifelike airway models for the simulated patients and generating realistic breathing curves based on the turbulence, velocity, and proportion of airflow through the nose and mouth. Each set of caregivers also had randomized, realistic breathing curves, and the researchers added standard hospital HVAC room ventilation to the model.
Accelerating computations with Microsoft Azure brought fast results
Strasser and Prichard used cloud computing to accelerate their CFD efforts, which let them perform the necessary three-dimensional computations and find answers in a reasonable timeframe. “Instead of having one gerbil running in a cage, you have thousands of them all running and communicating with each other to complete a task,” Strasser explains.
He notes, though, that more isn’t necessarily better with CFD. “Using twice as many machines to run the model doesn’t give you results twice as fast. There are diminishing returns, and we conducted a lot of testing to find the sweet spot between the number of Azure HBv2 virtual machines we needed to achieve the speed we required and keeping costs at a minimum.”
To put the magnitude of the research into perspective, it took 90 seconds for the airflow simulation model to achieve a quasi-steady-state solution. It took the research team 30 days of runtime, with eighteen 120-core Azure HBv2 virtual machines running 24 hours a day, to achieve those 90 seconds of flow time. “If we’d done this onsite, it wouldn’t have completed for decades,” says Strasser.
Their next step in the CFD simulation was to apply a mesh-blocking technique that breaks a volume of air (in this case, the hospital room) into smaller sections and then reconstructs it to define the entire space. “This model required 130-million mesh elements,” says Prichard. A more typical simulation is about a hundred times smaller. “To calculate results on a mesh that size, having such large-scale computational resources in Azure and the HBv2 machines was invaluable.”
Helping hospitals reduce airborne transmission of contagions
The CFD model correctly identified where exhaled particles moved throughout the hospital room and the FELIX-1 plastic face mask captured 95.8% of particle mass, compared to 88% with a surgical mask. Significantly, the team found that very few of the small number of particles that did escape (just 24 parts per billion) reached caregivers’ airways. These results will be significant as hospitals evaluate the FELIX-1 apparatus’s usage and effectiveness, not only pertaining to COVID-19 but also other contagions.
What’s next? Strasser and Prichard plan to create additional CFD models that incorporate real-life scenarios such as coughing and sneezing and evaluate different placements of the mask around the mouth and nose.
“Even though the critical phase around COVID-19 has passed, pneumonia and the flu are still leading causes of death,” says Prichard. “Better understanding of the effectiveness of this apparatus could substantially reduce airborne transmission in hospitals. And even a small decrease in transmission translates to thousands of lives saved each year. It’s exciting to know that we are helping hospitals with our research.”
Dr. Strasser is very optimistic about using Azure and a cloud-based HPC system to work on future models. “This experience totally changed my mind about using Azure. We had incredible support from that team, who very quickly set up a completely functional system for us. Better yet, the next models we run will take 1/10th of the time, and we will be able to use what we’ve built as a springboard to study new mask designs and how fluctuations in suction and cannula flow impact the airborne particles.”
“It was a confluence of the urgent need to run COVID-19 studies and the Microsoft grant availability. This put us in the unique position to become the only group in the world studying this specific device.”
Dr. Wayne Strasser, Associate Professor of Mechanical Engineering, Liberty University
Follow Microsoft