Researchers at Massachusetts General Brigham Hospital, in partnership with US-based chipmaker NVIDIA, have created an artificial intelligence (AI) model to predict oxygen needs of Covid-19 patients admitted to the emergency room.
The model is designed to determine if a patient requires supplemental oxygen hours or days following an initial exam.
Developed by Dr Quanzheng Li at Mass General Brigham, the original model called CORISK leverages medical imaging and health records to aid clinicians in managing hospitalisations.
To create a generalised model for multiple hospitals, NVIDIA and Mass General Brigham launched the EXAM federated learning initiative with 20 hospitals globally.
Researchers at individual hospitals used NVIDIA Clara Federated Learning Framework, along with a chest X-ray, patient vitals and lab values to train a local model.
The researchers then shared a subset of model weights with the global model as part of a privacy-preserving approach named federated learning. Data was obtained from patients across North and South America, Canada, Europe and Asia.
NVIDIA noted that the AI model is intended to predict the possibility that an individual showing up in the hospital emergency room will require supplemental oxygen.
This could help physicians decide the appropriate care level for patients, including admission to ICU.
Meanwhile, large-scale federated learning projects are ongoing to improve drug discovery and facilitate AI benefits at the point of care.
Owkin has partnered with NVIDIA, King’s College London and other organisations on a drug-discovery consortium called MELLODDY in the UK, to demonstrate the use of federated learning techniques in pharmaceutical industry.