AI4COVID-19
AN ARTIFICIAL INTELLIGENCE POWERED APP FOR DETECTING COVID-19 FROM COUGH SOUND
A project of
in collaboration with
&
Media Coverage
AI4COVID-19 has been reported by numerous media outlets around the globe. Below is a list of selected media coverages:
- Voice of America
- FOX News
- IEEE Spectrum
- Voice of America (Urdu version)
-
PTV Global
What is it?
Inability to test at scale has become Achille’s heel in humanity’s ongoing war against COVID-19 pandemic. A cost-effective screening/testing method, deployable at a global scale, can act as a game changer in this war.
To address this need, our multidisciplinary team of AI scientists and medical researchers has developed AI4COVID-19 app.
The app will be able to screen for COVID-19 infection, mainly from cough sound, in less than a minute, anytime, anywhere for anyone who has access to a smartphone and internet.
How it works?
Intuitive Explanation:
Audible differences in cough sounds have been used by physicians for differential diagnosis of respiratory infections for centuries.
Studies show that COVID-19 causes certain pathomorphological alterations in human respiratory system that are distinct from healthy subjects and several other viral and bacterial infections. These distinct pathomorphological alterations lead to certain distinct, albeit extremely subtle, features in cough sound.
This concept can be explained using a crude analogy of a classic guitar. If you place tiny objects of different size, shape and/or material composition inside the guitar one at time, it will lead to subtly different sound for different object, even when same notes are played. An appropriately trained ear may be able to guess the object inside the guitar by repeatedly listening to the notes corresponding to different objects.
However, the differences in cough caused by COVID-19 are so latent that human ear cannot differentiate these. Furthermore, the part of human body that produces cough sound (that includes the respiratory system) is way more complex than a guitar and is subject to population-based anatomical and other disease based physiological and pathological variations.
This makes detecting COVID-19 infection from cough an extremely complex multidisciplinary problem.
Our large multidisciplinary team set on solving this problem by leveraging years of prior experience in AI based diagnosis of respiratory diseases from cough sounds.
AI4COVID-19 works by sensing for the particular latent features in cough sound linked with distinct pathomorphological alterations in the respiratory system caused by the COVID-19 virus. Watch the explanatory video on the right to get the feel of how it works.
Explanatory video of AI4COVID-19
Technical Explanation:
The figure below shows the schematic of the AI engine that powers this app.
Simplified details of how it works can be found in our pioneering research paper on this idea. The current version of app uses much more intricate technology than the version reported in the paper. Prototype of the app has already been tested in the lab.
Inspired by our team's seminal paper, now many organizations around the world are trying to produce a similar app, some in collaboration with us and others independently.
Audible differences in cough sounds have been used by physicians for differential diagnosis of respiratory infections for centuries.
Studies show that COVID-19 causes certain pathomorphological alterations in human respiratory system that are distinct from healthy subjects and several other viral and bacterial infections. These distinct pathomorphological alterations lead to certain distinct, albeit extremely subtle, features in cough sound.
This concept can be explained using a crude analogy of a classic guitar. If you place tiny objects of different size, shape and/or material composition inside the guitar one at time, it will lead to subtly different sound for different object, even when same notes are played. An appropriately trained ear may be able to guess the object inside the guitar by repeatedly listening to the notes corresponding to different objects.
However, the differences in cough caused by COVID-19 are so latent that human ear cannot differentiate these. Furthermore, the part of human body that produces cough sound (that includes the respiratory system) is way more complex than a guitar and is subject to population-based anatomical and other disease based physiological and pathological variations.
This makes detecting COVID-19 infection from cough an extremely complex multidisciplinary problem.
Our large multidisciplinary team set on solving this problem by leveraging years of prior experience in AI based diagnosis of respiratory diseases from cough sounds.
AI4COVID-19 works by sensing for the particular latent features in cough sound linked with distinct pathomorphological alterations in the respiratory system caused by the COVID-19 virus. Watch the explanatory video on the right to get the feel of how it works.