Radiologists at Mass General are using AI to begin to analyze lung damage data from COVID-19 to predict the best treatment for patients.
Nvidia DGX A100 accelerators are helping the task, which involves using X-ray images of lungs to be combined with radiology data from other clinical insights to predict outcomes for COVID patients, according to an Nvidia blog.
Mass General Brigham used its own data to build the models. Once validated, they could be deployed in a hospital setting to track patient progress and offer treatment insights. Matthew D. Li, a radiology resident at Mass General and a member of the Martino Center QTIM Lab said there’s information in radiologic images not available to doctors as they make treatment plans.
“Using deep learning, we developed an algorithm to extract a lung disease severity score from chest X-rays that’s reproducible and scalable,” he said. That information can be tracked over time along with vital signs, pulse oximetry data and blood test results.
The Martinos Center is using the Nvidia DGX-1 to accelerate its work and is installing Nvidia DGX A100 systems, each with eight A100 Tensor Core GPUs that deliver 5 petaflops of AI performance. Such computational resources are essential to the work under the guidance of the Center for Machine Learning.
And MRI comparison AI is used with chest x-rays, with that training done on the DGX system. The center’s model labels 20 structures in a high-res X-ray and aligns them between two studies, requiring less than a second for the inference work. Another risk assessment model assigns a score for lung disease severity. Training was done on a public dataset over more than 150,000 chest X-rays and a few hundred COVID-positive X-rays from Mass General.
The severity score AI is under use by four research groups at the hospital with the Nvidia Clara Deploy SDX, which can be expanded for use beyond COVID to pulmonary edema.
Nvidia has listed its AI resources and work with healthcare institutions on its web site.