Skip to content
FIND A HEALTH VALLEY ACTOR
Embedded Systems Laboratory

COVID-19 Testimonial Series | Coughvid

07.09.2020
Share this article

How AI applied to cough recordings can help in the identification of COVID-19 cases

 

The following article is part of BioAlps’ testimonial series and was written by Mr Tomas Teijeiro, Scientist at the Embedded Systems Laboratory at the EPFL. Our series aims to provide a platform for the different life sciences actors from western Switzerland, who are active in finding and developing solutions to fight against the new coronavirus, to share their experience.

 

Presentation of Coughvid’s COVID-19 related activities

The Embedded Systems Laboratory at EPFL started the COUGHVID project, an initiative to study the potential of artificial intelligence applied to cough recordings to help in the identification of COVID-19 cases. The first stage of the project consisted on data collection, for which we followed two parallel strategies:

  1. a crowdsourcing approach, in which we collected more than 19’000 recordings through a multi-platform web application
  2. and collaboration with hospitals and pulmonologists to perform a clinical validation and expert annotation of part of the collected data, which resulted in 2’700 fully annotated cough recordings.

At this moment, we are fully involved in the curation and analysis of the database to publish a high quality, open dataset that may boost the research in this area.

 

Exogenous impact of COVID-19 on the project

As an academic institution, we are used to communicating our results via scientific publications, and only sometimes after a relevant result has been obtained, we explain it to the general public through the media. However, in this case, the interest of the media has been constant from the beginning, and we had to explain the scientific processes almost in real time.

Luckily, the COUGHVID initiative attracted a lot of attention from the media and other academic and industrial institutions. This was a key factor in the success of the crowdsourcing strategy for data collection and, in the end, it will determine our ability to test the initial hypothesis and the impact of the research outputs. Regarding the relationship with other partners, we noticed an increased aim for collaboration and the feeling that the relevance of our work is immediately put into value, which is particularly exciting!

Our work will be in principle the same, but for sure following COVID-19, all of us have learned important lessons about communication and collaboration, particularly efficient ways to communicate only relying on videoconferencing and online tools for the development of complete and complex research projects (like COUGHVID).

 

Endogenous impact of COVID-19 on the project

Usually, preparing a research project takes a lot of time and many iterations and discussions with the different partners to reach a convergence point, in a process that may take around 4-6 months. In this case, due to the tight timeline needed to address the pandemic, the process was shortened to just a couple of weeks, and some technical processes that could take up to a month were solved in a few hours. The engagement of all the participants and the feeling of urgency and importance of the tasks were crucial to achieving this.

Of course, we are learning many important lessons from this experience and, hopefully, it will help us to improve the efficiency of some processes. But I think the characteristics of this project are so special that probably (and  expectantly!) the future will be closer to our previous “normality” than to what we have experienced these last months.

 

Cantonal and/or Federal support measures | Coughvid’s feedback

The project is being developed by the lab members mainly as an overtime activity, and the few funds we required were provided by the ESL-EPFL own resources, so we didn’t rely at all on the specific COVID-19 related cantonal and federal support.

In the context of the complete society recovery these measures are important in a pandemic, but we should not forget that SARS-CoV-2 is a new strain of a family of viruses that have been already known by the health authorities since 2002, when the first SARS outbreak occurred. Therefore, in the context of research, these measures would be much more effective if additional funding for large-scale multi-disciplinary research projects are provided with enough time in advance.

Unfortunately, large amounts of funding cannot compensate the time needed to develop a complete research idea (and hopefully a complete solution) to similar types of pandemics.