A Swiss startup detects Covid-19 coughs

How Asthma Guardian used Youtube to test its AI and pivoted its app to monitor patients’ cough

What if there was an app that’s able to quantify Covid-19 cough? And thereby help to collect data for the treatment of the disease? Well, that’s exactly what Asthma Guardian is working on at the moment. 

Developed for asthma patients in the first place, Asthma Guardian’s AI can also recognize the cough of Covid-19 patients, as first tests have shown. The app is still in an experimental stage and not officially approved by any medical institution, but so far results are promising. 

The team behind Asthma Guardian started the research on Covid-19 hand in hand with Martin Brutsche, the chief doctor at the clinic for pneumology and sleep medicine at the cantonal hospital in St. Gallen. He’s convinced that Asthma Guardian could be useful for the treatment of lung diseases: “So far, it has been very difficult for us doctors to quantify coughs. But it’s important to objectify it for the success of medication and interventions”, the expert says. 

Emerging applications like Asthma Guardian that deliver information about a patient’s condition are closely watched in the industry. Michael Barczok, spokesman for the Federal Association of Pneumologists in Germany and head of the Ulm Lung Center, treats patients with lung diseases on a daily basis - currently via video call. He sees “great potential” for apps and other technologies that collect data, especially in his area of ​​expertise. 

The three founders of Asthma Guardian, Peter Tinschert, Iris Shih and Filipe Barata, carried out one of the biggest research projects ever related to asthma in Switzerland, together with the support of the CSS Health Lab, which is a cooperation between the ETH Zurich, the University of St. Gallen and CSS Insurance. Asthma Guardian is currently financing itself with a convertible loan and the first seed round is planned for the beginning of 2021.

This is how Asthma Guardian works for asthma patients: They download the app to their smartphone and keep the device next to their bed at night. The app monitors symptoms while the patient is sleeping and alerts the patient if his or her asthma needs further attention. 

How Youtube videos helped testing the AI 

In the middle of march, right when the Covid-19 pandemic hit globally, Peter Tinschert and his co-founders built their AI into an app for the first time - ready to test it after several years of researching. 

At the same time, first studies on Covid-19 showed that cough is one of the main symptoms of the infectious disease. So it became clear to the team that their AI can make a contribution to research and, eventually, to treatment by quantifying the course of patients’ cough. 

But first of all they had to overcome a hurdle: Asthma Guardian had only trained the AI on study data from asthma patients, they didn’t know how well it would get along with real world data. It was unclear, for example, whether the AI recognizes cough in general or only cough from asthmatics. Due to the lockdown though, the team was unable to carry out systematic tests. 

In that situation Youtube was of big help: People with various diseases recorded themselves when coughing and uploaded it on the platform. So Peter Tinschert was able to use these videos to test the AI and then he came to the conclusion: “If the AI ​​can recognize coughs from other diseases within videos with rather poor quality, then we can not only use it for asthma but also on Covid-19 patients."

Asthma Guardian in action 

Based on this knowledge, the team started a research project together with Martin Brutsche in St. Gallen. Three weeks later, Asthma Guardian’s AI was at work in the cantonal hospital. This process normally takes months, but in the wake of the pandemic, approving institutions including the Ethics Committee acted quickly, Peter Tinschert explains. 

Mobile phones with the app got attached to a few Covid-19 patients’ beds to monitor their breath and cough in order to collect as much data as possible about the development of their condition. The conclusion was promising. “The app has proven to be very reliable in a clinical setting. Practically all measurements could be used”, Martin Brutsche says.

If an app processes the patient’s information professionally, the doctor can get a good impression of the patient's condition without a personal meeting. The hope is that in areas where coverage by pneumologists is poor, time-consuming journeys to a doctor’s office far away could be replaced with online consultation hours, explains Michael Barczok, spokesman for the Federal Association of Pneumologists in Germany. 

At the same time apps could help to monitor Covid-19 patients outside the hospital as well. “It’s an advantage that the app is able to observe the course of the disease very well, so we can take those affected to the hospital early - and not only when it may be too late”, says Michael Barczok. 

Diagnosing Covid-19 through AI? 

Early research shows that it is possible to use AI for cough diagnostics, but Peter Tinschert thinks it’s too early to say that Covid-19 can be diagnosed through the technology. 

“Our AI has an accuracy of over 99%, but errors still occur. For diagnostics, especially in the case of infectious diseases such as Covid-19, AI errors carry a completely different risk”, he points out. Using the example of some Covid-19 antibody tests, the team could see that even those usually very precise tests may not be accurate enough. “My guess based on previous studies is that an AI will be inferior to these tests”, says Peter Tinschert. 

Big things ahead 

Asthma Guardian is still in a development stage, but according to Martin Brutsche, the self-measurement cough can be used as a biomarker for the disease activity of asthma, COPD (Chronic obstructive pulmonary disease), diseases of the lung structure and infections of the upper and lower respiratory tract – besides helping Covid-19 patients. The situations in which cough measurement with Asthma Guardian is particularly suitable have yet to be proven in studies, he says.