Facebook AI and NYU Build ML Models to Predict COVID Patient Needs
Researchers from Facebook AI and New York University (NYU) have developed three machine learning (ML) models that could help to predict how a COVID-19 patient's condition may develop, with the aim of helping hospitals manage their resources.
"Even a year into the pandemic, it remains challenging for doctors to predict how a patient's condition may change over the course of the disease," the Facebook AI blog post reads. "Will the patient improve in the next few days or worsen to the point where more intensive care is needed? With resources under unprecedented strain, it's important that hospitals know whether patients are likely to need escalated treatment and plan accordingly."
The new models were developed through a collaboration with NYU Langone Health's Department of Radiology and Predictive Analytics Unit. They include, 1) a model for predicting patient deterioration based on a single X-ray, 2) a model for predicting patient deterioration based on a sequence of X-rays, and 3) a model for predicting how much supplemental oxygen a patient might need based on a single X-ray.
Other approaches to apply ML to this matter have relied on "supervised training" and used single timeframe images. Supervised training models learn by training through a learning algorithm (linear regression, random forests, neural networks), which typically works through an optimization routine to minimize a loss or error function.
The process is well-defined in a 2020 research paper ("Supervised Machine Learning Models for Prediction of COVID-19 Infection using Epidemiology Dataset"):
Supervised learning techniques are ML learning techniques or algorithms that bind previous and current dataset with the help of labeled data to predict future events. The learning process begins with a dataset training process and develops targeted activity to predict output values. The techniques are able to provide results in input data with an adequate training process and compare results with actual results and expectations to identify errors and modify the model according to the results.
Supervised training methods are improving, the blog allows, but points out that labeling data is a time-consuming process.
"We chose instead to pretrain our ML system on two large, public chest X-ray data sets (MIMIC-CXR-JPG and CheXpert)," the blog post reads, "using a self-supervised learning technique called Momentum Contrast (MoCo). This allowed us to use large amounts of non-COVID chest X-ray data to train a neural network that could extract information from chest X-ray images. Then we fine-tuned the MoCo model using an extended version of the NYU COVID-19 data set."
This self-supervised learning approach--not having to rely on labeled data sets--is essential to this process, the researchers say, given that few research groups have enough COVID chest X-rays to train AI models. "Building AI models that can use a sequence of X-rays for prediction purposes is particularly valuable, because this method mirrors how human radiologists work, as using a sequence of X-rays is more accurate for long-term predictions. Importantly, this method also accounts for the evolution of COVID infections over time."
In their paper ("COVID-19 Deterioration Prediction via Self-Supervised Representation Learning and Multi-Image Prediction"), the researchers concluded:
… We found that our single-image prediction model was able to match a previous model… using separate advances, opening up new possibilities for research on synergies between contrastive pretraining and model architecture design. Our multi-image model performance surpassed that of all single-image models. In comparison to radiologists, our multi-image prediction model was comparable in its ability to predict patient deterioration and stronger in its ability to predict mortality. We hope these contributions will assist the community going forward on the task of hospital resource planning.
There was an emphasis in the announcement that this technology was designed to help physicians and medical professionals with their decision making, not replace them.
The paper is available online, and there's more on the Facebook AI blog.
John K. Waters is the editor in chief of a number of Converge360.com sites, with a focus on high-end development, AI and future tech. He's been writing about cutting-edge technologies and culture of Silicon Valley for more than two decades, and he's written more than a dozen books. He also co-scripted the documentary film Silicon Valley: A 100 Year Renaissance, which aired on PBS. He can be reached at firstname.lastname@example.org.