AI analysis of COVID patients' chest CT scans predicts ventilator need with 84% accuracy: study

Throughout the COVID-19 pandemic, hospitals and other healthcare facilities in the U.S. have been plagued by a seemingly never-ending stream of shortages: of personal protective equipment, ICU beds, ventilators, nurses and, now, oxygen.

The ventilator shortages that made headlines in the earliest months of the pandemic were eventually mitigated by increased manufacturing and a drop-off in new COVID cases thanks to the widespread availability of the three FDA-authorized vaccines. But as a slowing vaccination rate and the spread of the highly transmissible delta variant send case counts ticking back up, many hospitals are once again facing a lack of not just ICU beds, but also the critical care staff trained to operate ventilators—essentially creating a ventilator shortage in spite of the nation’s renewed stockpile of breathing support devices.

A new artificial intelligence tool, however, may provide the key to helping hospitals navigate these scarcities. Researchers from Case Western Reserve University developed an algorithm that they say can predict which COVID patients will require a ventilator with remarkable accuracy.

RELATED: FDA authorizes NASA-designed ventilator for COVID-19

A recently published study describes how the researchers trained the tool using initial chest CT scans of more than 600 patients diagnosed with the coronavirus in 2020. The deep learning AI was able to identify certain characteristics that couldn’t be detected by the naked eye in the scans of patients who ended up needing respiratory support in the ICU, differentiating them from those who didn’t require a ventilator.

After the algorithm had been trained, it was tasked with analyzing another set of about 250 chest CT scans from patients in the U.S. and Wuhan, China. In that retrospective test, the AI predicted which patients had ultimately needed a ventilator with 84% accuracy.

The AI algorithm could therefore help hospitals not only determine how many ventilators they’ll need but also be more precise and proactive as they plan out each COVID patient’s treatment, according to Anant Madabhushi, Ph.D., one of the authors of the study.

“This tool would allow for medical workers to administer medications or supportive interventions sooner to slow down disease progression,” said Amogh Hiremath, a Case Western graduate student in Madabhushi’s lab and lead author of the study.

“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,” Hiremath said.

RELATED: Medtronic deploys remote-controlled ventilators to lessen coronavirus exposure

Next up, Madabhushi said, the research team is hoping to deploy their AI for real-time use in a handful of local Cleveland hospitals.

Clinicians at University Hospitals and Louis Stokes Cleveland VA Medical Center would then be able to upload a CT scan to the cloud-based tool and quickly receive a prediction of whether the patient in question will require a ventilator.