Covid-19 Artificial Intelligence Diagnosis Using Only Cough Recordings(ieeexplore.ieee.org)
ieeexplore.ieee.org
Covid-19 Artificial Intelligence Diagnosis Using Only Cough Recordings
https://ieeexplore.ieee.org/document/9208795
3 comments
Built a data collection pipeline of COVID-19 cough recordings through our website (opensigma.mit.edu) between April and May 2020 and created the largest audio COVID-19 cough balanced dataset reported to date with 5,320 subjects.
Thanks a lot for posting. Are you involved with the project? If so, great work, and I am excited that you made transfer learning work for this task!
I am interested if you have given the following suggestions some thought: 1) The number of false positives is quite high, which is a challenge for a screening tool. However, the app could be used as a precursor to a quick antigen test: Rapidtests have basically no false positives, and your model basically has no false negatives. 2) The model could in principle be extended to be able to detect that a user's cough has changed. The user would record his "healthy" coughs as a baseline, and the model would try to detect a change when the user gets Covid-19. In principle this approach has the advantage that it is a intra-subject comparison, so the model could deal better with the particularities of a particular user (their mic, their voice and so on). Of course gathering the required training data might be a bit harder, since you now need two recordings from each user/patient.
The paper mentions that this model is currently being tested in several hospitals. How quickly do you think it could be rolled out?
I am interested if you have given the following suggestions some thought: 1) The number of false positives is quite high, which is a challenge for a screening tool. However, the app could be used as a precursor to a quick antigen test: Rapidtests have basically no false positives, and your model basically has no false negatives. 2) The model could in principle be extended to be able to detect that a user's cough has changed. The user would record his "healthy" coughs as a baseline, and the model would try to detect a change when the user gets Covid-19. In principle this approach has the advantage that it is a intra-subject comparison, so the model could deal better with the particularities of a particular user (their mic, their voice and so on). Of course gathering the required training data might be a bit harder, since you now need two recordings from each user/patient.
The paper mentions that this model is currently being tested in several hospitals. How quickly do you think it could be rolled out?
The one thing I would like most in the world, is to see the IEEE panels...in person.