Limited availability of testing kits for COVID-19. Test kit may not be accurate : https://edition.cnn.com/2020/04/28/health/coronavirus-antibody-tests-terrible/index.html. Test kit result needs long time to be produced : https://www.cnet.com/how-to/how-long-will-the-coronavirus-test-take-and-when-will-i-get-my-covid-19-results/. Not all hospitals or medical facilities is equipped with advanced medical imaging such as CT Scan. Human has limit on radiation intake from X-Ray, we can not take chest X-ray frequently. Point-of-care ultrasound (POCUS) is expensive and small medical facilities especially in remote rural third world countries mostly don't have it. In the other AI/ML hackathon, my AI enabled stethoscope had received feedback that is not suitable for remote diagnostic and social distancing policy. In the other AI/ML hackathon, my AI/ML to identify specific signature of COVID-19 from lung voice recording, had received feedback not all people can create their own DIY less than 1$ digital stethoscope.
What it does
Using a mobile phone (including the very very cheap Android phone such $10 LG Lucky), we can find the specific vocal signatures that could indicate someone is infected with COVID-19. Starting with a web app which can be opened by the minimum Android Webkit to capture voice recording.
How I built it
https://s2.smu.edu/~eclarson/pubs/2011UbiComp_Cough.pdf has proven that low cost microphone can be used to sense chronic obstructive pulmonary disease (COPD). https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=6987276 shows that pneumonia can be detected from wavelet features of the cough sounds. https://ijisrt.com/an-integrated-computerized-cough-analysis-by-using-wavelet-for-pneumonia-diagnosis points that neural network is able to distinguish the pneumonia with other respiratory diseases. https://link.springer.com/chapter/10.1007/978-3-662-49381-6_29 highlights cough detection on low-cost mobile devices in real-time has high degree of accuracy.
Challenges I ran into
I am still learning Flutter. With Flutter it can be transform into a multi-platform mobile app.
Accomplishments that I'm proud of
With Google CoLab, I can create the prototype.
What I learned
I am lucky enough, at least, I can have some data. My neighbor who lives couple of blocks away, has been declared heal from COVID-19, still keeps the recordings in the phone.
What's next for Coughing AI
Getting more recording of COVID-19 patients.
Try It out
google-colab, google-tensorflow, php, python