AI in Healthcare: Detecting COVID-19

AI4COVID-19 allows testing for COVID-19 via a smartphone app. This new AI technology could change how we respond to some respiratory illnesses.

Image by pch.vector on Freepik

Detecting and diagnosing infectious diseases can be a difficult task for our healthcare services, especially in the midst of a pandemic. Artificial intelligence technologies are a promising solution to this problem.

A dry cough is one of the most common symptoms of a COVID-19 infection. Coughing is also a key way that the virus is spread from person to person. So, an inexpensive and widely available diagnostic tool utilizing coughs would be beneficial for reducing the spread of such a disease.

That’s exactly what the researchers and developers behind AI4COVID-19¹ set out to do.

The Problem With Current Diagnosis Solutions

Testing for COVID-19 at a large scale was a struggle for our healthcare systems during the peak of the pandemic. In New Zealand, we had two main types of tests: the RAT (rapid antigen test) and the PCR (polymerase chain reaction) tests.

To get a PCR test, you typically had to line up at a testing station somewhere in your neighborhood. RAT tests were made available later in the pandemic and allowed people to test for COVID-19 without having to leave the house. In some parts of the world, PCR test kits could be delivered to your home for self-collection and then sent back to the lab for testing. There are a few problems with each of these testing methods.

Cost to patients. In NZ, PCR and RAT tests were available for free if you went through the public health system. Otherwise, RAT tests were being sold for up to $18.99² per single-use test. In some other parts of the world, there were no free RAT test options and you had to pay for a test each time you needed one.

Cost to organizations. During the peak of the pandemic, there were cues for miles at each testing station, with thousands of people waiting to be tested. Many were symptomless but wanted to be tested just to be safe. For each PCR test, swabs are taken and stored in a sterile environment, DNA fragments called primers are needed to amplify the genetic material being tested, and PCR machines need to run multiple cycles per individual test. That’s a lot of testing equipment being used, especially when millions of people all over the world are being tested. The cost of this adds up pretty quickly.

Availability of testing. Test kits (whether PCR or RAT), were not always available to everyone. Some who live in remote areas relied on telehealth appointments during the pandemic and did not have access to local testing stations.

Increased exposure to the virus. In-person testing increased the risk of exposure to the virus. Although this problem was mostly solved by the introduction of at-home RAT tests, these still had to be purchased or picked up in-store.

Time to receive results. With the initial PCR testing, it could take a few days to receive the results of your test. Again, this was improved with the introduction of RAT tests. RAT test results are available within 15 minutes. However, false negatives are common with RAT tests¹.

User errors. At-home tests, such as the RAT test, are prone to false negatives by user error. To ensure a RAT test will give an accurate result, the nasal swab has to pick up enough material. This means putting the swab high enough in your nostril and moving it around for the correct amount of time.

Physical waste. Both PCR and RAT tests create a significant amount of physical waste. For every test, there are swabs, plastic wrappings, tubes, and cassettes. For tests done in a lab, there is even more waste per test. Things like pipette tips and Eppendorf tubes can only be used once. Although there are good reasons why all of this testing equipment is single-use, it does cause an awful lot of waste.

The Solution: A COVID-19 Detection App

AI4COVID-19 is a solution for all of these problems. It is an app that records a cough and determines whether that cough is likely to be caused by COVID-19 or some other respiratory illness.

How can this be possible?

One of the most common symptoms of COVID-19 is a dry cough; approximately 67.7% of those who have had COVID-19 experienced a dry cough³. When a person is infected with COVID-19, one of the key ways they can spread it to other people is through coughing⁴.

Although not everyone who is COVID-19-positive experiences a cough, those who do are at a higher risk of spreading it to others. By this logic, it just makes sense to have some kind of cough-detection and diagnosis method. However, a cough is also a symptom of many other respiratory illnesses, which makes this rather difficult.

Each respiratory illness affects the body in a specific location and nature that is unique to that disease. These differences can cause the coughs associated with each disease to have different traits. COVID-19 is no exception. The SARS-CoV-2 virus causes pathomorphological changes (how the virus affects the body) distinct from other respiratory illnesses⁵ ⁶. AI-based analysis of CT scans⁷ and x-rays⁸ ⁹have also confirmed these differences.

Because of these pathomorphological differences, the cough associated with COVID-19 has different traits than other respiratory illnesses. However, unaided ears cannot reliably distinguish between these differences.

That’s where AI4COVID-19 comes in.

AI4COVID-19 uses trained domain-aware AI to detect the specific traits in a COVID-19 cough. ‘Domain-aware’ means that instead of solely relying on blind big data churning, the AI engine relies on the deep domain knowledge of medical researchers who specialize in infectious and respiratory diseases¹.

Essentially, medical researchers set out to analyze the pathomorphological changes caused by the COVID-19 virus and determine whether these would result in a cough with different traits to other respiratory illnesses. As mentioned above, researchers found that this is, in fact, the case.

Using a domain-aware AI design decreased the amount of time and data needed to accurately test the hypothesis. This was important because time is a luxury that researchers did not have in the midst of a pandemic.

How AI4COVID-19 works

To use AI4COVID-19, all you need to do is open the app and record a cough. Each cough recording is 3 seconds long. The recording is then sent to the AI server in the cloud where it goes through two testing stages: cough detection and cough diagnosis.

The cough detection engine will determine whether the recording actually contained a cough. It is trained to distinguish the sound of a cough from 50 different kinds of environmental noise, so it can even be used in public areas where you are unlikely to have a silent background. If there is no cough detected, the user is prompted to re-record the cough. If a cough is detected, the recording is then sent to the diagnosis engine.

The diagnosis engine uses three different classifier systems:

• First-class classifier: Deep Transfer Learning-based Multi Class classifier (DTL-MC)
• Second class classifier: Classical Machine Learning-based Multi Class classifier (CML-MC)
• Third class classifier: Deep Transfer Learning-based Binary Class (DTL-BC)

The results of each of these classifier systems are then sent to the mediator. If the systems return identical outcomes the user is given either a positive or negative result. If they are not identical, the test is inconclusive. Results are displayed to the user in about 2 minutes.

There are three possible outcomes to be displayed on the app:

• ‘COVID-19 likely’ — if all classifier systems return a positive result
• ‘COVID-19 unlikely’ — if all classifier systems return a negative result
• ‘Inconclusive’ — if all classifier systems do not return identical results

How accurate is this app?

Currently, this technology has only been tested with a small amount of data; however, it is promising.

The accuracy of each system within the AI engine is displayed in the table below.

Image by Author

How does this solve the current problems?

Although AI4COVID-19 is not a clinical-grade tool, it provides the ability to screen for COVID-19 on a large scale, without many of the problems of current testing methods. This technology may also be applied to other respiratory illnesses in the future, so the solutions it provides are not COVID-19-specific.

The table below covers the problems mentioned earlier in this article and how AI4COVID-19 provides the solution.

Image by Author

Conclusion

Although AI4COVID-19 is not a clinical-grade tool, it is a promising advance in technology that could solve many of the problems with current large-scale testing methods. This solution is not COVID-19-specific either. The cough-screening AI engine could be applied to other respiratory illnesses in the future.

References

1. Imran A, Posokhova I, Qureshi HN, et al. AI4COVID-19: AI enabled preliminary diagnosis for COVID-19 from cough samples via an app. Inform Med Unlocked. 2020;20:100378. doi:10.1016/j.imu.2020.100378
2. Keane J. Big price differences on rapid antigen tests. Consumer. March 2, 2022. Accessed December 30, 2022. https://www.consumer.org.nz/articles/big-price-differences-on-rapid-antigen-tests
3. World Health Organization. February, 2020. Report of the WHO-China joint mission on coronavirus disease 2019 (COVID-19) 2020. https://www.who.int/docs/default-source/coronaviruse/who-china-joint-mission-on-covid-19-final-report.pdf
4. Cohen J. Not wearing masks to protect against coronavirus is a ‘big mistake,’ top Chinese scientist says. Science. March 27, 2020. Accessed December 30, 2022. https://www.science.org/content/article/not-wearing-masks-protect-against-coronavirus-big-mistake-top-chinese-scientist-says
5. Bai HX, Hsieh B, Xiong Z, et al. Performance of Radiologists in Differentiating COVID-19 from Non-COVID-19 Viral Pneumonia at Chest CT. Radiology. 2020;296(2):E46-E54. doi:10.1148/radiol.2020200823
6. Tian S, Hu W, Niu L, Liu H, Xu H, Xiao SY. Pulmonary Pathology of Early-Phase 2019 Novel Coronavirus (COVID-19) Pneumonia in Two Patients With Lung Cancer. J Thorac Oncol. 2020;15(5):700–704. doi:10.1016/j.jtho.2020.02.010
7. Li L, Qin L, Xu Z, et al. Using Artificial Intelligence to Detect COVID-19 and Community-acquired Pneumonia Based on Pulmonary CT: Evaluation of the Diagnostic Accuracy. Radiology. 2020;296(2):E65-E71. doi:10.1148/radiol.2020200905
8. Narin A, Kaya C, Pamuk Z. Automatic detection of coronavirus disease (COVID-19) using X-ray images and deep convolutional neural networks. Pattern Anal Appl. 2021;24(3):1207–1220. doi:10.1007/s10044–021–00984-y
9. Wang L, Lin ZQ, Wong A. COVID-Net: a tailored deep convolutional neural network design for detection of COVID-19 cases from chest X-ray images. Sci Rep. 2020;10(1):19549. Published 2020 Nov 11. doi:10.1038/s41598–020–76550-z