Wheezing, crackling, and rumbling are signals that a respiratory disease is underway. To diagnose lung conditions, clinicians use a stethoscope to decipher lung sounds and spirometry to measure how much air a person can breathe in and out. However, subtle wheezing is easy to miss, and spirometry is difficult to use practically, especially in children.

Dohyeong Kim, a public policy researcher at the University of Texas at Dallas, and his collaborators at Seogyeong University developed a real-time wheeze counting algorithm to differentiate normal and abnormal breath sounds and track their frequency (1). Clinicians could use this technology to remotely diagnose, monitor, and treat respiratory conditions, while patients could use it to self-manage symptoms. 

What motivated you to conduct this study?

My background is in public and environmental health research from both the social science and medical perspectives. I have used Geographic Information Systems for spatial modeling of health and environmental data for most of my career. In 2018, I started working with medical doctors and computer scientists to integrate clinical data from lung patients with environmental indicators. We published one study showing an association between lung functionality and levels of air pollutants using a deep learning model (2). A major drawback of that research was our inability to collect lung sound data in real time. We published this latest study as pilot evidence that we can count how many abnormal sounds occur within a certain period. We developed the AI tools to diagnose symptom onset and symptom exacerbation through continuous monitoring of lung sounds. This is just one of several papers we are working on. We plan to eventually integrate lung sound data with environmental indicators, such as particulate matter levels and ozone concentration, to provide personalized and timely alerts to people who may be exposed to adverse levels of environmental pollutants. 

What are the advantages of your system's predictions for patients with respiratory conditions?

A known trigger for asthma is indoor and outdoor air pollution. There are systems in place to monitor outdoor ambient air pollutants well, but not indoor pollutants. However, modern populations stay longer indoors. It would be beneficial for people to monitor their exposure to pollutants when they cook inside or when they light the fireplace so that we can collect those data and make a correlation with lung functionality. We aim to identify patterns that may not pose a concern to the general population but may be life threatening for individuals with pre-existing respiratory issues like asthma or chronic obstructive pulmonary disease. 

Our system identifies subtle wheezing better than a human. Doctors can detect subtle sounds with a stethoscope, but sometimes you want immediate feedback to decide if you need to go to a hospital. 
- Dohyeong Kim, University of Texas at Dallas

Severe wheezing is easily noticeable, but mild breathing difficulties are harder to identify. Our system identifies subtle wheezing better than a human. Doctors can detect subtle sounds with a stethoscope, but sometimes you want immediate feedback to decide if you need to go to a hospital. We want to predict potential asthma attacks and facilitate early intervention.  

What would be the challenge of implementing this system for monitoring lung illnesses?

Our AI modeling algorithm works well. The data comes from a source that is widely used in many papers because it is already labeled, which helps train the algorithm. Labeling new patient data is challenging because medical professionals are busy, and sometimes they are not willing to spend their time on it. As the algorithm becomes smarter, there will be less need for intensive intervention by a human doctor. All the doctors we have been working with think that this is really promising, and they are willing to use it, but we do not know how much it will be used. 

Doctors currently lack the tools for real-time health monitoring and counting wheezing events. The hope is that more doctors will see the benefits of using our system — especially its ability to offer real-time insights — and will increasingly incorporate it into their practice. 

What is next for this project?

We have obtained funding to develop a lung sound patch (3). Once it is fully developed and implemented, we will be able to put it on the chests of lung patients, where it can capture lung sounds in real time, filter out noise, and store the data in the cloud. Then using our algorithm, the system will automatically count wheezing events and provide doctors with data that allows them to give patients timely warnings or medical advice. This should facilitate data collection for improving the algorithm, which has been difficult to do because we need to obtain patient consent, and sometimes patients are in critical condition. Once this patch is available, data collection will be more convenient because the device is small and unobtrusive. We're also planning to combine this with any other types of biomarkers, such as body temperature or heart sounds, that people are willing to share. 

This interview has been condensed and edited for clarity.

References

1. Im, S. et al. Real-time counting of wheezing events from lung sounds using deep learning algorithms: Implications for disease prediction and early intervention. PLOS ONE   18, e0294447 (2023).
2. Kim, D., Cho, S., Tamil, L., Song, D. J. & Seo, S. Predicting Asthma Attacks: Effects of Indoor PM Concentrations on Peak Expiratory Flow Rates of Asthmatic Children. IEEE Access  8, 8791–8797 (2020).
3. Lee, S. H. et al. A wearable stethoscope for accurate real-time lung sound monitoring and automatic wheezing detection based on an AI algorithm. (In Review, 2023). doi:10.21203/rs.3.rs-2844027/v1