Facebook builds A.I. to predict likelihood of worsening Covid symptoms

Artificial intelligence researchers at Facebook claim they have developed software that can predict the likelihood of a Covid patient deteriorating or needing oxygen based on their chest X-rays.

Facebook, which worked with academics at NYU Langone Health’s predictive analytics unit and department of radiology on the research, says that the software could help doctors avoid sending at-risk patients home too early, while also helping hospitals plan for oxygen demand.

The 10 researchers involved in the study — five from Facebook AI Research and five from the NYU School of Medicine — said they have developed three machine-learning “models” in total, that are all slightly different.

One tries to predict patient deterioration based on a single chest X-ray, another does the same with a sequence of X-rays, and a third uses a single X-ray to predict how much supplemental oxygen (if any) a patient might need.

“Our model using sequential chest X-rays can predict up to four days (96 hours) in advance if a patient may need more intensive care solutions, generally outperforming predictions by human experts,” the authors said in a blog post published Friday.

William Moore, a professor of radiology at NYU Langone Health, said in a statement: “We have been able to show that with the use of this AI algorithm, serial chest radiographs can predict the need for escalation of care in patients with Covid-19.”

He added: “As Covid-19 continues to be a major public health issue, the ability to predict a patient’s need for elevation of care — for example, ICU admission — will be essential for hospitals.”

In order to learn how to make predictions, the AI system was fed two datasets of non-Covid patient chest X-rays and a dataset of 26,838 chest X-rays from 4,914 Covid patients.

The researchers said they used an AI technique called “momentum contrast” to train a neural network to extract information from chest X-ray images. A neural network is a computing system vaguely inspired by the human brain that can spot patterns and recognize relationships between vast amounts of data.

The research was published by Facebook this week but experts have already questioned how effective the AI software can be in practice.

“From a machine learning perspective, one would need to study how well this translates to new, unseen data from different hospitals and patient populations,” said Ben Glocker, who researches machine learning for imaging at Imperial College London, via email. “From my skim reading, it appears that all data (training and testing) is coming from the same hospital.”

The Facebook and NYU researchers said: “These models are not products, but rather research solutions, intended to help hospitals in the days and months to come with resource planning. While hospitals have their own unique data sets, they often don’t have the computational power necessary to train deep learning models from scratch.”

“We are open-sourcing our pretrained models (and publishing our results) so that hospitals with limited computational resources can fine-tune the models using their own data,” they added.