Micro Focus is now part of OpenText. Learn more >

You are here

You are here

How AI can be a COVID-19 game-changer

David Weldon Freelance Writer and Research Analyst

With most US states now reporting sustained increases in new coronavirus cases, fear about the pandemic's resurgence is on the rise. That is placing renewed pressure on the key elements in this healthcare battle, including early detection, containment, triage and diagnosis, and vaccine development.

Artificial intelligence (AI) has the potential to bring an arsenal of important weapons to this fight. But the reviews are mixed on how effectively healthcare has used AI in the past. Many experts hope that the current crisis will change that.

Consider: AI can allow faster predictions about the spread of the virus and potential hot spots. It can help healthcare workers more quickly and accurately diagnose cases. It can empower public health experts to quickly measure the effectiveness of defensive steps taken to slow the disease spread. And it can aid in the effective distribution of emergency resources.

In recent years, AI might have earned a reputation as a buzzword, with many technology companies hyping up its presence in their products, said Irma Rastegayeva, co-founder and chief innovation catalyst at eViRa Health.

"But despite this reputation and valid ethical concerns, AI's capabilities and power hold real promise for healthcare, especially when it comes to finding ways to understand, control, and eventually stop the coronavirus pandemic." 
Irma Rastegayeva

Here's how AI could be a game-changer in the battle against COVID-19.

Devouring droves of data

The most obvious benefit that AI can bring to the fight against COVID-19 is its ability to handle large, complex datasets, said Charles Parker, a consultant specializing in the use of AI in healthcare, financial services, and the energy sector.

"The analytics can get very complex for predicting populations that move around. This is where AI can bring significant compute capabilities and do the 'what if's' that we cannot easily do with human computing."
Charles Parker

The difference here, though, is that AI must be taught the rules and may not immediately understand how a thing like changing weather may impact the models, he said. This would require new training or rules knowledge. However, once trained, "AI can do these calculations and predictions 24 hours a day across large datasets." 

That tireless analysis can enable smarter decisions about therapy availability, critical supply manufacture and distribution, the allocation of healthcare workers, and the management of all of these processes, Parker said.

The prominent role of machine learning

Of all the various technologies that make up what is collectively referred to in the singular as "artificial intelligence," perhaps the one with the greatest potential benefit in the fight against COVID-19 is machine learning. (Machine learning combines large amounts of data with fast, iterative processing and intelligent algorithms that allow the software to learn automatically from patterns or features in the data.) 

Part of the reason why machine learning has such great potential is the large number of organizations already invested in it.

Nearly every health organization is using machine learning in some capacity, according to Deborah DiSanzo, an instructor of AI in health at Harvard University's T.H. School of Public Health. DiSanzo was previously the general manager of IBM Watson Health from 2015 to 2018.

It may be as simple as using natural-language processing to transcribe physicians' and radiologists' notes, or as advanced as matching patients to medicines based on the DNA of their particular disease, for the benefit of its patients, physicians, or the health system as a whole.

"Every academic medical center, every national cancer institute, and every large health system has started, partnered, or developed a machine-learning pilot."
Deborah DiSanzo

The rapid speed in which health organizations have been able to identify the disease, track the spread, repurpose existing medicines, identify potential vaccines, triage patients, and assign clinicians and equipment to a particular floor, unit, or hospital can be attributed, in some way, to machine learning, DiSanzo said.

As good news as all of this is, DiSanzo is quick to note that there is also a lot of misinformation about what AI can and can’t do in this fight. She said there are four major areas where healthcare professionals need to improve decision making.

Early detection and epidemic analysis

There has been much written over the past few months exclaiming that AI models are able to diagnose COVID-19 from X-Ray or CT images. "This is all overdone," DiSanzo said.

Someday, machine-learning models will be able to read CT and X-ray images to determine that pneumonia is present with a high sensitivity and specificity, she said. And someday, these models may be able to distinguish sufficiently between different types of pneumonia.

For now, however, all these models need more data. They need a variety of images across a number of genetic makeups, and they need sufficient quantity of both normal images, images of patients with COVID-19 pneumonia, and images of patients with different types of pneumonia, she added.

"[When this data is put into the models] and we have sufficient sensitivity and specificity above 90%, we can say that the models are diagnosing COVID-19 pneumonia. Now, they are pilots and experiments, only."
—Deborah DiSanzo


Location tracking on mobile phones is very effective at knowing where we are. Containment is just one step from that. Models, in some countries, are signaling when infected people enter a building, a train, or a store.

In some countries, entry is being restricted based on location tracking and these models, DiSanzo said. Additionally, companies are installing thermal imaging cameras to take employees' temperatures before they enter a building. This, combined with surveys and vital signs from wristbands, are determining if employees can work, and can stay at work.

Containment using location and vital signs is not a technical issue, DiSanzo said. Technically, this can be done. But it's a societal and personal liberty issue. Mobile phone apps provide sufficient information to track us today. The question is, 'Do we want this information used to contain the COVID-19?'"

Triage and diagnosis

This is one of the strongest use cases for machine learning for COVID-19. Temperature, heart rate, respiration, and blood oxygen are vital signs that can be taken with a high degree of accuracy through non-invasive, remote sensors, DiSanzo said.

Other health signs—including voice, movement, weight, and toileting—can also be effectively monitored with remote sensors. Combined, machine-learning models are being created to monitor and predict the onset of adverse events and the progression of disease, DiSanzo said.

These systems are advanced and work well, she said. They will help clinicians care for patients in their own homes, in nursing homes, in hospitals, and in extended care facilities.

"These remote sensors and models will ensure that we are taking care of patients in the most appropriate facility, and that the most vulnerable patients are identified quickly and monitored effectively."
—Deborah DiSanzo


Models have been and are being developed to understand the novel coronavirus DNA, RNA, and protein structure. A variety of machine-learning techniques will continue to help scientists in their discovery of vaccines, DiSanzo said. 

Time for an assist

We are at the very beginning of how machine learning can help us in a pandemic. And we have learned with COVID-19, that healthcare needs to be better, faster and more efficient, DiSanzo said.

"We are seeing early successes in pilots. We must have the will to stay for the long run. And when we see success, it will not be that the machines replace epidemiologists, or radiologists, or chemists or clinicians. It will be that machine learning has helped all these professions do their jobs better, faster and more efficiently."
—Deborah DiSanzo

Keep learning

Read more articles about: Enterprise ITData Centers