The study used a machine-learning technique called federated learning (FL) to create an algorithm using data from chest x-rays and electronic health data such as vital signs and laboratory test results from patients with COVID-19 symptoms. Researchers called the FL model electronic medical record chest x-ray AI model (EXAM).

The goal of the research was to build an AI tool that predicts how much extra oxygen a COVID-19 patient may need in the first days of hospital care. EXAM was trained on data from 16,148 patients in hospitals across five continents.

The input to EXAM included 20 demographic and clinical features, such as age, white blood cell count, respiratory rate, and blood pressure. The output corresponded to the intensiveness of the patient's oxygen therapy status at the 24- and 72-hour periods following initial admission to the emergency room. The oxygen therapy categories included room air, low-flow oxygen, high-flow oxygen, noninvasive ventilation, and mechanical ventilation, as well as patient death.

Federated learning, also known as collaborative learning, is a machine-learning technique that trains an algorithm across multiple decentralized devices holding local data samples. Federated learning differs from traditional, centralized machine-learning techniques, where all of the data are uploaded to a single server. A major advantage of the FL approach in a medical setting is that the lack of centralization eliminates regulatory hurdles associated with data sharing, security, anonymity, and privacy.

The FL approach allowed researchers in the EXAM project to maintain strict patient confidentiality. Only the algorithm, not the data, was shared across hospital sites. EXAM represents one of the largest and most dispersed clinical FL studies to date.

"Usually in AI development, when you create an algorithm on one hospital's data, it doesn't work well at any other hospital," said first author Dr. Ittai Dayan, PhD, of Massachusetts General Hospital Radiology and Harvard Medical School, in a statement. "By developing the EXAM model using federated learning and objective, multimodal data from different continents, we were able to build a generalizable model that can help frontline physicians worldwide."

Following initial training, EXAM was tested at three different hospitals in Massachusetts: Cooley Dickinson Hospital, Martha's Vineyard Hospital, and Nantucket Cottage Hospital. The results showed that EXAM predicted the oxygen needed within 24 hours of a patient's arrival in the emergency room with an overall sensitivity of 95% and an overall specificity of over 88%.

The researchers were also able to show that the federated approach provided a 16% improvement in predictive accuracy measured across all participating sites and an average increase in generalizability of 38% when compared with models trained at a single site using that site's data.

"Federated learning has transformative power to bring AI innovation to the clinical workflow," said co-author Dr. Fiona Gilbert, an honorary consultant radiologist at Addenbrooke's Hospital and chair of radiology at the University of Cambridge School of Clinical Medicine. "Our continued work with EXAM demonstrates that these kinds of global collaborations are repeatable and more efficient so that we can meet clinicians' needs to tackle complex health challenges and future epidemics."

Hardware company Nvidia contributed to EXAM, and the project's code is being hosted on Nvidia's site.

Do you have a unique perspective on your research related to artificial intelligence? Contact the editor today to learn more.

Machine learning predicts if a COVID-19 clinical trial will be successful
Using machine learning, researchers have teased out the underlying factors of success for COVID-19 clinical trials, according to a PLOS One article...

COVID-19 therapies could come from existing drugs
At least four drugs that have already been approved by the U.S. Food and Drug Administration may be effective as COVID-19 therapies, according to a study...