How Artificial Intelligence Helps Us Fight Pandemics

The novel coronavirus outbreak has quickly become the largest pandemic in recent history, but it’s not unprecedented. The outbreak of the so-called “Spanish Flu”, an avian influenza virus, spread worldwide, infecting one-third of the population. While scientists are still learning how the coronavirus operates, we have lots of tools at our disposal to fight it.

In the world of ever-growing datasets, artificial intelligence can make connections that even the most diligent of scientists miss. While they have a lot to learn before making insights, I believe AI can save the world.

Visualization of the coronavirus (via Fusion)

Faster diagnoses

Identifying positive coronavirus cases is critical for mitigating its spread, but that’s hard to do.

In the US, a shortage of COVID-19 tests means that many more people have the virus than what we’re counting for. It also means that we’re likely underestimating the number of fatalities from the virus. We’re seeing shortages of the vital chemicals needed to carry out these tests, and a lack of production to keep up with demand. This is in part because creating actual tests is a very complicated process, you can’t exactly make these at home. Even if you could, scientists aren’t sure how accurate these tests are

With tests in short supply, and cases still on the rise, some scientists are looking to other ways to diagnose coronavirus. In places where testing isn’t available, chest X-rays and CTs have been found to accurately diagnose coronavirus. While results from testing can take hours to days, X-rays and CTs are immediate. In China’s Hubei Province, CT scans became an effective tool for diagnosing coronavirus quickly, and importantly, they can show how intensely an infection is progressing. Radiographic scans can reveal “opacities in the lungs, essentially spots where fluid is building up, or where the tissue is damaged.

A CT image of a 36-year-old male COVID-19 patient with bilateral ground‐glass opacities in their lungs  (Bernheim et al, 2020)

While a person could look through scans and identify features indicative of COVID, pneumonia, or other respiratory infections, an AI can do it faster, with more precision, and can identify patterns across a large dataset. Looking at datasets of CT scans from patients with and without the virus, AIs have been able to make correct diagnoses 0.95 area under the curve (this is a scaling system that measures the accuracy, with 1 being the most accurate, 0 being the least). While these studies have looked at small datasets, and while we still have a lot to learn about COVID-19, these studies show promising results that AIs can identify COVID more accurately than traditional tests.

Individualized treatments

In addition to diagnosing infections, artificial intelligence tools can also accurately predict the severity of the virus in patients. Using data from a small sample of patients at two Chinese hospitals, an algorithm was able to track factors leading to which patients would eventually develop fatal fluid build-ups in their lungs. Surprisingly, their model found that common COVID symptoms were not useful in predicting which patients would experience severe symptoms. In contrast to other studies, neither age or gender seemed to play a role either (it is important to remember that this study looked at patients already hospitalized, so this might look different than the general population)

The AI found that the highest risk factors for patients were their hemoglobin levels, ALT levels (an enzyme important for liver function), and muscle aches were the most definitive warning signs of worsening conditions.

By knowing the warning signs and risk factors, doctors can better allocate their limited resources, hopefully leading to lower fatalities and more recoveries.

As cases continue to rise, internet platforms track the global spread (via Isaac Quesada)

Using science to inform policy

With artificial intelligence, its always important to remember: all algorithms are programmed by people, and the results you see are dependent on the code you write. Every study is different, and each has its own set of assumptions and caveats–be careful of the results you see in the news.

In a time where things are moving fast, scientific discoveries just keep coming and coming. With that in mind, it’s important to consider that quantity does not always equal quality. In the coming years, studies will be refined, methods improved, and theories disproved.

The best we can do is work with the information we have. At the very least, we need to listen to scientists.

–Lissie Connors