Add to favorites

#Industry News

The (Sober) State of Artificial Intelligence in the Fight Against COVID-19

If you ask us at The Medical Futurist about the importance of artificial intelligence in healthcare, we will have a lot to talk about.

We’ve seen how it could solve alarm fatigue in hospitals. We’ve analyzed the unusual associations the technology discovered in medicine. We believe it will usher the real era of the Art of Medicine. Dr. Meskó even embarked on a journey to better understand the language of A.I.

So of course, with the COVID-19 pandemic, we had to explore the contribution of A.I. in this public health crisis. We came across promising endeavours involving such algorithms from mining for insights through tracking the spread to detecting the infection via coughs.

It’s sensible to be vigilant and to know about the potential privacy issues and dangers that come with the use of such new technologies in unprecedented times. However, these developments we elaborate upon below will help us better prepare for future outbreaks.

A.I. predicting outbreaks

In what is now a quasi-prophetical tale, it was an A.I. company that issued the first warnings of an outbreak. BlueDot used its algorithm to sift through hoards of news reports, airline data, and reports of animal disease outbreaks to detect trends. These were then analyzed by epidemiologists who then alerted the company’s clients. The software even correctly predicted the virus’ likely path from Wuhan to Tokyo after it first appeared.

As the disease spread, other organizations turned to similar solutions. A team of researchers fed an algorithm with anonymized air travel and smartphone movement data to explore how the disease could spread from Wuhan to other cities soon after it appeared. Another team used an A.I. to model COVID-19’s spread from case reports, human movement and public health interventions. This helped show how travel restrictions hampered the contagion’s growth.

By employing such methods, authorities can get a better insight into forthcoming disease outbreaks and better prepare for any eventuality.

Aiding in diagnosis with radiological scans, facial analysis and… coughs!

Identifying those positive with COVID-19 is of urgency to stop cross-contamination. However, with overburdened healthcare institutions, identifying the infected proves to be challenging. Hospitals are overflowing with lines of patients showing up for various ailments, most of whom don’t need additional care. A.I. can help in the screening and triage of relevant patients in these cases and alleviate the pressure on hospitals.

In China, the Zhongnan Hospital used an A.I. software programme to detect signs of pneumonia associated with SARS-CoV-2 infections on images from lung CT scans. This helps assist radiologists to screen patients and prioritize potential COVID-19 cases for further testing.

Teaming with Microsoft, Providence launched an online tool to distinguish those who could have been infected with COVID-19 from those with less threatening conditions. In its first week, over 40,000 patients used the tool. Following suit is Partners HealthCare with its A.I. COVID-19 Screener chatbot to screen people remotely.

Tampa General Hospital in Florida went for a more sci-fi-like option. It deployed an A.I. system in collaboration with Care.ai to detect feverish (and potentially COVID-19 positive) visitors via facial scan.

A group of researchers and engineers from San Francisco launched an initiative called Cough for the Cure. Yep, you’ve guessed it: they aim to develop a COVID-19 diagnostic tool based on cough audio. Once enough recordings from people tested positive or negative for the illness have been collected, it will be the task of a machine learning algorithm to detect nuances to differentiate between those people and aid in diagnosis.

Resource management and predicting severe outcomes

Lack of protective equipment, shortage of hospital beds and overloaded ICU are worldwide phenomena these days. With A.I.-based forecasting tools, hospitals can better manage their resources.

Qventus developed a software programme aimed to help hospital administrators during the pandemic. Their model takes into consideration the patient influx from COVID-19 and the related deaths, and forecasts its effect on the hospital’s capacity like beds, ICU and ventilator capacity.

Researchers published their findings on creating an A.I. framework to assist in rapid clinical decision-making. Their predictive models use real patient data to determine who will develop acute respiratory distress syndrome (ARDS), a severe complication in COVID-19. These models achieved 70% to 80% accuracy in predicting severe cases. Thus, patients identified in this way could be prioritized for specialized support.

These two methods tie in to better allocate and prioritize resources in due time, and to ultimately lessen the burden on healthcare institutions.

Speeding up vaccine research

In a move to put medical research around COVID-19 on the fast track, organizations like Microsoft and the Allen Institute for A.I. set up a free, comprehensive database with over 29,000 related scientific articles. Believed to be the most extensive collection of research papers on the topic, this COVID-19 Open Research Dataset (CORD-19) aims to facilitate research work by being an easily accessible resource for scientists. Moreover, it allows machine learning algorithms to mine for insights which can aid towards crucial research, be it for vaccine development, transmission trends or even unusual associations.

Combining machine learning and network science, the BarabasiLab is looking for new drug candidates against the novel coronavirus. Less than 10 days since the team refocused its Network Medicine toolset for this purpose, they already have a list of drugs to be tested in human cell lines in an experimental lab.

For the skeptics

As we mentioned in the introduction, all of these examples are not solutions but rather directions for future use. It’s not all praise for A.I. There are reasons to be skeptical about its help against the pandemic in the same way there are reasons to be optimistic about it.

No one could predict the scale of COVID-19, so solutions are only popping up now. This means that we lack prior data specific to this contagion which is paramount to A.I.’s functioning. For instance, some years ago Google launched Flu Tracker in order to predict flu outbreaks. The tech giant put a stop to the project in 2013 after failing to predict that year’s peak by 140%. This was partly attributed to a lack of trustworthy data. Now newer software can scour from a much wider range of sources. As we gather more pertinent data, developers can build more robust algorithms around this outbreak and even better prepare for future ones.

Undoubtedly, the measures involved to deploy such solutions raise a whole new level of privacy concerns. We’ve seen governments from South Korea, Singapore and Israel employing phone surveillance to track COVID-19’s spread. Germany is also contemplating a similar option. In the U.S., controversial companies like Clearview AI, which used its A.I. technology to build a facial identification database from social media photos for its clients, are contenders to bring such surveillance tools for the government’s use.

Inflating the capabilities of A.I. will lead to unrealistic expectations and ill-informed investments, which can lead to the crash of the A.I. industry. As such, a level of caution is healthy, but the technology’s potential to support health crises remains undeniable.

We need all the help we can get

It’s important to note that in these settings, A.I. alone is not the solution. Rather, such algorithms should be considered as an aid for professionals. In the case of BlueDot for instance, the algorithm reported on trends it identified by going through scores of data which would take a considerably longer time by a human. Once these trends were found, it was human epidemiologists who analyzed them to give a definite verdict.

Ongoing and future progress made in the medical A.I. field will become undeniable assets for healthcare practitioners beyond the pandemic. Yes, efficacy and privacy issues are valid and require a concerted effort from policymakers, authorities and the general public for secure and efficient use of A.I. in fighting health crises; because let’s have no doubt, we won’t be able to tackle the next pandemic without implementing A. I. way before it appears. And we all stand to gain from the contribution of A.I. in healthcare.

The (Sober) State of Artificial Intelligence in the Fight Against COVID-19

Details

  • Avenue André Roussin, Marseille, France
  • The Medical Futurist