Challenge
Producing a system to detect indications of COVID-19, including in patients with less-pronounced symptoms, without additional, cost-prohibitive tests / unavailable testing.
UC San Diego Health’s Solution
Although CT scans can be used to detect signs of COVID-19, it would be impractical and costly to perform a CT scan on every patient presenting with symptoms such as fever and fatigue. UC San Diego Health—deploying a machine learning-based algorithm that two of its radiologists had already been working on in the context of chest x-ray pneumonia detection using publicly available chest x-rays—was able to quickly create an experimental procedure for detecting COVID-19 signs through chest x-rays. “Their algorithm generates a probability map, which is a heat map that gets overlaid on top of the chest x-ray image,” says Mike Hogarth, clinical research information officer at UC San Diego Health.
Benefits of Using AWS
When the COVID-19 pandemic began, UC San Diego Health already had a HIPAA compliant Amazon Web Services (AWS) environment. We began discussing with AWS the prospect of developing the process for containerizing the algorithm project and had a scalable infrastructure in an aggressive 10 days timeline. With this infrastructure, a gateway puts x-ray images into Amazon Elastic Compute Cloud (Amazon EC2) instances running on Docker; the machine learning algorithm analyzes them and applies the heat map overlays, and then the images travel back to be viewed by clinical personnel.
“I want to underscore the benefit of having a HIPAA-secure environment is a highly scalable cloud infrastructure and the benefit of having this gateway—in just 5 days, we were able to implement a real clinical workflow connected to the real clinical PACS imaging system,” says Dr. Hogarth UC San Diego Health, Clinical Research Information Officer.
Panelists:
