Dr. Dan Ponticiello, 43, and Dr. Gabriel Gomez, 40, intubates a patient with coronavirus disease (COVID-19) at the COVID-19 ICU at Providence Mission Hospital in Mission Viejo, California, January 8, 2021.
Lucy Nicholson | Reuters
Facebook’s artificial intelligence researchers claim they have developed software that can predict the likelihood that a Covid patient will deteriorate or need oxygen based on their chest X-rays.
Facebook, which worked with academics from NYU Langone Health’s predictive analytics unit and the radiology department for the study, says the software could help doctors avoid sending at-risk patients home too early, while also helping hospitals planning the oxygen demand.
The 10 researchers involved in the study – five from Facebook AI Research and five from the NYU School of Medicine – said they developed a total of three machine-learning “models,” all of which differ slightly.
One tries to predict the patient’s deterioration from a single chest X-ray, another does the same with a series of X-rays, and a third uses a single X-ray to predict how much supplemental oxygen (if any) a patient will need. .
“Our model using sequential chest X-rays can predict up to four days (96 hours) in advance of whether a patient may need more intensive care solutions, generally better than the predictions of human experts,” the authors said in a Friday. published blog post.
William Moore, professor of radiology at NYU Langone Health, said in a statement, “ We have been able to demonstrate that using this AI algorithm, serial chest X-rays can predict the need for escalation of care in patients with Covid-19. . “
He added, “Since Covid-19 is still a major public health concern, the ability to predict a patient’s need for increased care – for example, ICU admission – will be essential for hospitals.”
To learn how to make predictions, the AI system received two datasets of non-Covid chest radiographs from patients and a dataset of 26,838 chest radiographs 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-rays. A neural network is a computer system vaguely inspired by the human brain that can recognize patterns and recognize relationships between vast amounts of data.
The research was published by Facebook this week, but experts have already wondered how effective the AI software can be in practice.
“From a machine learning perspective, you should examine how well this translates into new, unseen data from different hospitals and patient populations,” said Ben Glocker, who researches machine learning for imaging at Imperial College London, via email. “My short lecture shows that all data (training and testing) come from the same hospital.”
The researchers from Facebook and NYU said, “These models are not products, but rather research solutions, designed to help hospitals with resource planning in the days and months to come. Hospitals have their own unique datasets, but they often lack the computing power they need. to train deep learning models from scratch. “
“We use our pre-trained models open source (and publish our results) so that hospitals with limited computing power can fine-tune the models with their own data,” they added.