AI tool predicts which COVID-19 patients need ventilator

The AI tool, developed at Case Western Reserve University, Ohio, analysed CT scans from nearly 900 COVID-19 patients diagnosed in 2020 and was able to predict ventilator need with 84 per cent accuracy.

“That could be important for physicians as they plan how to care for a patient - and, of course, for the patient and their family to know,” said Anant Madabhushi, the Donnell Institute Professor of Biomedical Engineering at Case Western Reserve and head of the Center for Computational Imaging and Personalized Diagnostics (CCIPD). “It could also be important for hospitals as they determine how many ventilators they’ll need.”

MORE ON ENGINEERING'S RESPONSE TO COVID-19

Next, Madabhushi said he hopes to use those results to try out the computational tool in real time at University Hospitals and Louis Stokes Cleveland VA Medical Center with COVID-19 patients.

If successful, he said medical staff at the two hospitals could upload a digitised image of the chest scan to a cloud-based application, where the AI at Case Western Reserve would analyse it and predict whether that patient would need a ventilator.

Among the more common symptoms of severe COVID-19 cases is the need for patients to be placed on ventilators but the number of such machines has often outpaced supplies, which led some hospitals to practise “splitting” ventilators, a procedure in which a ventilator assists more than one patient.

To date, doctors have lacked a consistent and reliable way to identify which newly admitted COVID-19 patients are likely to need ventilators, information that could prove invaluable to hospitals managing limited supplies.

Researchers in Madabhushi’s lab began their efforts to provide such a tool by evaluating the initial scans taken in 2020 from nearly 900 patients from the US and from Wuhan, China where the first known cases were diagnosed.

In a statement, Madabhushi said those CT scans revealed, with the help of AI, distinctive features for patients who later ended up in the intensive care unit and needed help breathing.

The research behind the tool is published in IEEE Journal of Biomedical and Health Informatics.

Amogh Hiremath, a graduate student in Madabhushi’s lab and lead author on the paper, said patterns on the CT scans could not be seen by the naked eye, but were revealed only by the computers.

“This tool would allow for medical workers to administer medications or supportive interventions sooner to slow down disease progression,” Hiremath said. “And it would allow for early identification of those at increased risk of developing severe acute respiratory distress syndrome - or death. These are the patients who are ideal ventilator candidates.”