Text and photo: Peter Witten, AAU Communication and Public Affairs

Every year, around three million people die from COPD - Chronic Obstructive Pulmonary Disease - worldwide. One of the diseases that claims the most lives, according to WHO.

Here in Denmark alone, over 3,000 COPD patients die annually. About 400,000 Danes have COPD, but many of them don't know it. Not yet. And the disease cannot be cured.

Depressing facts, but still, COPD patients can be helped.

Three students from the Department of Computer Science at Aalborg University want to contribute to this with a new research project. Using sound recordings and artificial intelligence/AI. The three are Tobias Oliver Eberhard, Cuno Sankey-Olsen og Rasmus Hvass Olesen.

Live longer

COPD patients often have had symptoms for years before the diagnosis is made. This is critical because the disease cannot be cured, only slowed down.

But the sooner COPD is detected and treatment starts, the better the quality of life and longer lifespan. 

"With early diagnosis, we have the opportunity to help patients with lifestyle changes and initiate treatment. And teach them how to best manage their disease and thus counteract disease progression," says professor and chief physician Ulla Møller Weinreich from the Clinical Institute at AAU and the Department of Pulmonary Medicine at Aalborg University Hospital.

The sound of COPD

Ulla Møller Weinreich has helped the three students with sources for their research. About the project, she says: "The idea is brilliant!"

The idea is to use AI to analyze voice samples from both COPD patients and from people who have not been diagnosed with the deadly lung disease. To investigate whether sound recordings can quickly detect COPD in those who do not yet know they have the disease.

"We can already now determine that AI can indeed see special things in voice samples from people with COPD," says Tobias Oliver Eberhard.

Cough three times

The three students have collected voice samples from nearly 100 individuals. Half of them have COPD, while the other half have not been diagnosed with COPD.

The voice samples consist of sound recordings of about one and a half minutes. The test subjects must read vowels, a short story, and finally cough three times. This provides enormous amounts of digital data, which are analyzed with artificial intelligence.

Just one second of sound recording provides about 50,000 lines of "scribbles" - data code.

The algorithm

The plan with the research project is not to develop a ready-to-use app that can immediately determine whether you have COPD or not.

"We focus on creating an algorithm that can hopefully contribute to further research. To confirm or refute whether it is possible to recognize COPD with the help of machine learning/AI," say Tobias, Cuno, and Rasmus.

"It would be great if we could develop something that can ultimately be used in apps, telemedicine equipment, or in hospitals," they add. 

Professor Ulla Møller Weinreich believes that AI analysis of voice samples could become a reality in healthcare: "I think it could be a good screening tool in future work with COPD, so we can identify those we need to examine further. Time will tell if the sensitivity and specificity of the test are sufficient to make it possible," she says.

Mega-motivating

The three students have been in contact with many COPD patients who wanted to provide voice samples: "They know that this project will not help them. But they all say that if it can help future generations, it means a lot."

For Tobias, Cuno, and Rasmus, it also means a lot if their research project can benefit others. "It is mega-motivating that our project can potentially make a difference," they emphasize.

Their supervisor on the project is Associate Professor Ilhan Aslan from Department of Computer Science, AAU.

He is convinced that there are great future opportunities in precisely the type of research that Tobias, Cuno, and Rasmus are engaged in.

"Super hearing"

"This is just one of several possibilities. There is great potential in using voice analysis to understand what is happening in the human body," he says.

Ilhan Aslan compares voice analysis using AI to when doctors use a stethoscope to listen to the human body. With the use of artificial intelligence, it suddenly becomes "super hearing."