Image: The DeepBreath AI algorithm uses deep learning to identify respiratory disease (Photo courtesy of EPFL)
The distinctive whooshing sound made by air traversing the intricate network of tiny lung passageways changes significantly when those channels are affected by asthmatic inflammation or blocked by bronchitis-associated secretions. The process of listening to these sound changes using a stethoscope—known as auscultation—is an indispensable part of nearly every health examination. Despite over 200 years of stethoscope usage, the interpretation of auscultation remains largely subjective, with different physicians often hearing varying sounds. The accuracy also varies based on the healthcare provider's experience and area of specialization. These complexities present an ideal opportunity for deep learning, which could offer a more objective interpretation of audio patterns. Deep learning has already proven its worth in augmenting human interpretation of complex medical tests like X-rays and MRI scans.
Now, a team of researchers at EPFL (Lausanne, Switzerland) and University Hospital Geneva (HUG, Geneva, Switzerland) has developed an intelligent stethoscope, Pneumoscope, that is powered by a novel AI algorithm - DeepBreath. The breakthrough tool holds promise in enhancing the management of respiratory diseases, especially in resource-limited and remote locations. The algorithm was trained using patient data from Switzerland and Brazil, and then validated using recordings from Senegal, Cameroon, and Morocco, thus offering insights into its geographic adaptability. In a study, the AI algorithm, DeepBreath, demonstrated how automated interpretation could revolutionize respiratory disease diagnosis. About 600 pediatric outpatients from five countries (Switzerland, Brazil, Senegal, Cameroon, Morocco) participated in the study, with the focus being on the three most common types of respiratory disease - pneumonia confirmed by radiography, and clinically diagnosed bronchiolitis and asthma. Despite the small patient cohort, DeepBreath demonstrated an impressive performance across various locations, signifying the potential for further improvement with more data.
A significant contribution of the study was the use of additional methods to understand the inner workings of the algorithm’s “black box”. The research team successfully showed that the model indeed relied on the breathing cycle for its predictions and highlighted the most important parts. By confirming that the algorithm genuinely uses breath sounds rather than biased signatures in the background noise—termed as "cheating"— the study fills a crucial void in current literature and increases confidence in the algorithm. The multidisciplinary team is now focusing on preparing the algorithm for real-world deployment in the Pneumoscope. The next substantial step involves repeating the study with a larger patient pool, using recordings from this newly developed digital stethoscope that also captures temperature and blood oxygen levels.
“Respiratory disease is the number one cause of preventable death in this age group,” explained Professor Alain Gervaix, Head of the Department of Pediatric Medicine at HUG. “This work is a perfect example of a successful collaboration between HUG and EPFL, between clinical studies and basic science. The DeepBreath-powered Pneumoscope is a breakthrough innovation for the diagnosis and management of respiratory diseases.”