We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

eess.AS

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Electrical Engineering and Systems Science > Audio and Speech Processing

Title: AI4COVID-19: AI Enabled Preliminary Diagnosis for COVID-19 from Cough Samples via an App

Abstract: Inability to test at scale has become Achille's heel in humanity's ongoing war against COVID-19 pandemic. An agile, scalable and cost-effective testing, deployable at a global scale, can act as a game changer in this war. To address this challenge, building on the promising results of our prior work on cough-based diagnosis of a motley of respiratory diseases, we develop an Artificial Intelligence (AI)-based test for COVID-19 preliminary diagnosis. The test is deployable at scale through a mobile app named AI4COVID-19. The AI4COVID-19 app requires 2-second cough recordings of the subject. By analyzing the cough samples through an AI engine running in the cloud, the app returns a preliminary diagnosis within a minute. Unfortunately, cough is common symptom of over two dozen non-COVID-19 related medical conditions. This makes the COVID-19 diagnosis from cough alone an extremely challenging problem. We solve this problem by developing a novel multi-pronged mediator centered risk-averse AI architecture that minimizes misdiagnosis. At the time of writing, our AI engine can distinguish between COVID-19 patient coughs and several types of non-COVID-19 coughs with over 90% accuracy. AI4COVID-19's performance is likely to improve as more and better data becomes available. This paper presents a proof of concept to encourage controlled clinical trials and serves as a call for labeled cough data. AI4COVID-19 is not designed to compete with clinical testing. Instead, it offers a complementing tele-testing tool deployable anytime, anywhere, by anyone, so clinical-testing and treatment can be channeled to those who need it the most, thereby saving more lives.
Subjects: Audio and Speech Processing (eess.AS); Machine Learning (cs.LG); Sound (cs.SD); Quantitative Methods (q-bio.QM); Machine Learning (stat.ML)
Cite as: arXiv:2004.01275 [eess.AS]
  (or arXiv:2004.01275v5 [eess.AS] for this version)

Submission history

From: Usama Masood [view email]
[v1] Thu, 2 Apr 2020 21:39:34 GMT (2423kb,D)
[v2] Mon, 6 Apr 2020 02:06:26 GMT (2536kb,D)
[v3] Tue, 7 Apr 2020 04:59:39 GMT (2499kb,D)
[v4] Thu, 16 Apr 2020 05:55:56 GMT (2754kb,D)
[v5] Tue, 5 May 2020 22:53:50 GMT (2602kb,D)

Link back to: arXiv, form interface, contact.