Artificial intelligence algorithms trained at EPFL to see and hear Covid in the lungs

Artificial intelligence and deep learning find more and more applications in the healthcare sector. Ultrasounds and sound recordings can, for example, be screened by AI models to diagnose Covid-19. It is for this use case that EPFL’s Global Health (iGH) smart group is developing the DeepChest and DeepBreath deep learning algorithms. The first specializes in lung imaging and the second in respiratory sounds. “Artificial intelligence allows us to better understand the complex patterns in these fundamental clinical exams. So far, the results are very promising, ”says Professor Martin Jaggi of the EPFL’s Faculty of Informatics and Communication.

Evolution of other pre-Covid projects

Both DeepChest and DeepBreath were born before the new coronavirus pandemic began. DeepChest began in 2019 and initially aimed to define markers that could better distinguish viral pneumonia from bacterial pneumonia. The clinical part of the project is currently underway at the CHUV (Lausanne), where thousands of lung ultrasound images of patients suffering from Covid-19 symptoms and admitted to the emergency room have been collected. DeepBreath was born from a project developed at HUG (Geneva) to better diagnose pneumonia. In this context, breathing sounds have been collected since 2017 to design an intelligent digital stethoscope, called a “pneumoscope”. These registrations are now used to develop the DeepBreath algorithm. Researchers hope it will be able to diagnose Covid-19 by the end of the year.

Fight against other respiratory diseases

“We continue to improve and validate algorithms and make it easier for doctors to interpret the complex logic of the black box. We want to create robust and reliable tools whose capabilities go beyond this pandemic, ”confides Mary-Anne Hartley, physician and researcher with the Global Health smart group. In addition to Covid-19, these algorithms promise to be useful in the fight against other respiratory diseases and also to meet the growing challenge posed by antibiotic resistance, the EPFL says.

Remember that other EPFL researchers are working on algorithms that can diagnose Covid-19 from the sound of a cough recorded on a smartphone.