Deep learning diagnostic and risk-stratification pattern detection for COVID-19 in digital lung auscultations: clinical protocol for a case–control and prospective cohort study

• DOI: 10.1186/s12890-021-01467-w
• Publication Date: 2021-03-24T13:06:08Z
• Authors: Alban Glangetas; Mary-Anne Hartley; Aymeric Cantais; Delphine S. Courv...; David Rivollet; Deeksha M. Shama; Alexandre Perez; Hervé Spechbach; Véronique Trombert; Stéphane Bourquin; Martin Jaggi; Constance Barazzo...; Alain Gervaix; Johan N. Siebert
• Keywords: Auscultation; Triage; Stethoscope

Lung auscultation is fundamental to the clinical diagnosis of respiratory disease. However, a subjective practice and interpretations vary widely between users. The digitization acquisition interpretation particularly promising strategy for diagnosing monitoring infectious diseases such as Coronavirus-19 disease (COVID-19) where automated analyses could help decentralise care better inform decision-making in telemedicine. This protocol describes standardised collection lung auscultations COVID-19 triage sites deep learning approach diagnostic prognostic modelling future incorporation into an intelligent autonomous stethoscope benchmarked against human expert interpretation.A total 1000 consecutive, patients aged ≥ 16 years meeting testing criteria will be recruited at screening amongst inpatients internal medicine department Geneva University Hospitals, starting from October 2020. diagnosed by RT-PCR on nasopharyngeal swab COVID-positive are followed up until outcome (i.e., discharge, hospitalisation, intubation and/or death). At inclusion, demographic data collected, age, sex, medical history, signs symptoms current episode. Additionally, recorded with digital 6 thoracic each patient. A algorithm (DeepBreath) using Convolutional Neural Network (CNN) Support Vector Machine classifier trained these audio recordings derive prediction (COVID positive vs negative) risk stratification categories (mild severe). performance this model compared baseline random subset sounds, blinded physicians asked classify audios same categories.This has broad potential standardise evaluation various levels healthcare, especially context decentralised monitoring.PB_2016-00500, SwissEthics. Registered April