Researchers at Massachusetts General Hospital’s Athinoula A. Martinos Center for Biomedical Imaging described today how they’re using Nvidia Corp.’s latest artificial intelligence systems to aid their fight against COVID-19.

The researchers are using AI to create new models that can segment and align multiple chest X-ray scan images, allowing them to get a better understanding of the damage the disease causes to a patient’s lungs over time. Armed with this information, clinicians will then be able to make better decisions regarding the patient’s treatment, the researchers said.

“While helping hospitalists on the COVID-19 inpatient service, I realized that there’s a lot of information in radiologic images that’s not readily available to the folks making clinical decisions,” said Matthew D. Li, a radiology resident at Mass General and member of the Martinos Center’s QTIM Lab. “Using deep learning, we developed an algorithm to extract a lung disease severity score from chest X-rays that’s reproducible and scalable — something clinicians can track over time, along with other lab values like vital signs, pulse oximetry data and blood test results.”

The Martinos Center said it’s using Nvidia’s most advanced DGX-1 AI platform, which is powered by the company’s latest A100 graphics processing units, to accelerate its research.

CT scans and X-rays are done on patients suffering from COVID-19 in order to check for damage in their lungs. By comparing these images with follow-up scans, clinicians can see if a patient is recovering from the disease or if the damage is getting worse.

One of the problems they face is that segmenting and lining up two scans that are taken from slightly different positions or angles is extremely difficult. But this is necessary because the changes can be almost imperceptible to the human eye.

“Radiologists spend a lot of time assessing if there is change or no change between two studies,” said Bruce Fischl, director of the Martinos Center’s Laboratory for Computational Neuroimaging. “This general technique can help with that. Our model labels 20 structures in a high-resolution X-ray and aligns them between two studies, taking less than a second for inference.”

Fischl said his AI models can also be used in conjunction with a risk assessment model that analyzes chest X-rays and assigns a score, known as the “Kalpathy-Cramer’s lung disease severity score,” to each patient.

The risk assessment model, trained on a public dataset of more than 150,000 chest X-rays plus several hundred COVID-positive X-rays from Mass General, provides a consistent and quantitative metric on the impact of the lung disease that can further aid clinicians.

In a separate research effort, Brandon Westover, executive director of Mass General Brigham’s Clinical Data Animation Center, has created AI models using systems that run on Nvidia’s Quadro RTX 8000 GPU. Those models can predict clinical outcomes for each patient diagnosed with COVID-19, he said.

The model works by analyzing 30 variables and creating a risk score for each patient that predicts whether a patient may end up needing critical care, as well as their likelihood of dying from the disease. For patients already admitted to hospital, a neural network is used to predict the hourly risk that they’ll need artificial breathing support within the next 12 hours, taking into account variables such as their age, vital signs, respiratory rate and other signs. It enables clinicians to see which patients are most at risk, and assign resources and plan treatments as necessary.

“These variables can be very subtle, but in combination can provide a pretty strong indication that a patient is getting worse,” Westover said.

Images: Massachusetts General Hospital