Diagnosing a patient’s condition from their coughing behavior is something that medical professionals have done since ancient times. Using AI to do this form of analysis is a huge challenge, and one that has appealed to the Embedded Systems Lab, since the outbreak of COVID-19.
To characterize a cough as a COVID cough, or as a tubercular cough, is a mammoth task. It requires the compiling of databases of authentic coughs of each type, and the training of deep neural networks to a point where they can associate a particular type of cough to a known illness.
We launched the CoughVid project during the outbreak of COVID-19, compiling an enormous database, and developing our technology accordingly.
It is also possible to track the number of coughs a patient suffers over a certain period, using a device that can sense coughs through sound and movement. The device in this case must distinguish between the coughs it is required to identify and yawns, laughs, and other gestures that have similar sounds and movements. Furthermore, in order to reassure patients of the discretion of the device, it must run autonomously, without connection to the Internet.
These tasks involve cutting edge technologies: low-power devices, sensing on the edge, DNNs, our X-HEEP platform and many other areas which are at the heart of the Embedded Systems Laboratory. They also present a fantastic opportunity to collaborate with industry. We are working with VersaSens and Sensemodi to build real-world demonstrators, and with the University Hospital of Lausanne (CHUV) to work with medical experts and volunteer patients.
Bonsai: cough monitoring device | |
Cough Benchmarks | |
Coughvid | |
TB Diagnosis | |
News:
Cough-E – Using edge technology to track coughs discreetly
Coughvid Crowdsourcing Dataset