A cervical patch that monitors breath sounds can help manage asthma and chronic obstructive pulmonary disease (COPD) by detecting symptom flare-ups in real time, without compromising patient privacy.
Asthma and COPD are two of the most common chronic respiratory diseases. In Europe, the combined prevalence is about 10% of the general population. In Canada, approximately 3.8 million people suffer from asthma and two million people suffer from COPD.
The chronic nature of asthma and COPD requires ongoing disease monitoring and management. Patients with these conditions share many similar clinical symptoms, such as frequent coughing, wheezing, and shortness of breath. These symptoms may worsen from time to time and from situation to situation, such as exposure to smoke.
Even with optimal treatment, patients encounter unpredictable flare-ups or exacerbation of their condition. These can become fatal and require immediate medical attention. Effective and predictive tools, which allow continuous remote monitoring and early detection of exacerbation, are essential for prompt treatment and better health.
An international collaboration between Canada and Germany with expertise in upper respiratory health, audio/acoustic engineering and wearable computing is developing a wearable device to monitor these respiratory symptoms.
Wearable technologies have been widely applied for remote monitoring of asthma and COPD. Most of these devices have built-in microphones to collect audible clinical symptoms, such as coughing, from patients. However, such designs hinder full patient compliance due to privacy concerns regarding continuous monitoring of all sounds in their daily encounters and home environment.
Efficient and intelligent algorithms are needed for wearable health devices to meaningfully interpret data as it enters the system. Recent advances in artificial intelligence (AI) have rapidly changed many areas of medical diagnosis and therapeutic monitoring.
However, the AI “black box” problem also creates ethical and transparency issues in biomedicine. Most AI tools only let us know the input and output of the algorithm (e.g., transforming an input X-ray image into a predicted output diagnosis), but not the processes and functionings in between. This means that we don’t know how AI tools do what they do.
Moreover, the implementation of real-time analyzes in portable devices is difficult due to the limited computational resources in these devices, but is essential for the rapid detection of respiratory tract symptoms. The development of a reliable and cost-effective “wearable AI” is crucial for this project.
To address these unmet challenges, our AI-powered wearables will have the ability to protect conversation privacy and perform near real-time data analysis to enable patients and clinicians to take informed action without delay. .
Listening with protected speech privacy
At McGill University, the Canadian team is developing a wearable device, similar in size to a Fitbit, to track and monitor upper airway health during daily activities. The device is based on mechanical-acoustic detection technology.
In a nutshell, a small patch-like skin accelerometer is customized to be placed on the neck. When a person experiences upper respiratory tract symptoms such as coughing, hoarse voice, etc., the body sounds characteristic of these symptoms create acoustic waves that travel to the skin of the neck and turn into mechanical vibrations detectable by the skin accelerometer.
Most recognizable speech features are in the high frequency range (about six to eight kilohertz). The tissues of the human neck serve as a filter that only the low frequency components of a signal can pass through. This means that identifiable voice information is detectable as sound by our sensors but inaudible to the human ear, thus preserving the privacy of users’ speech.
We are currently working on the development of a smartphone application that will connect to the portable device. This mobile app will generate an upper respiratory health diary summary for patients. Additionally, with the consent of the users, the report can also be sent to their primary healthcare providers for remote monitoring.
Small and smart AI
At the Friedrich-Alexander-Universität Erlangen-Nürnberg, the German team has developed deep neural networks, a specific subfield of AI, which are very lightweight and require very little computing memory of less 150 kilobytes. In addition, continuous monitoring generates a large and complex data source. In a recent post, we reported that our algorithms are on par with state-of-the-art algorithms, even if they fit on a low-cost microcontroller.
Our current project will build on these findings and extend these cost-effective AI algorithms to automate the analysis of mechanical acoustic signals. This information, along with other user-specific data (such as local air quality and painkiller used), can be used to predict the risk of exacerbation of asthma/COPD symptoms in a patient.
Currently, the device is in the testing phase. By examining the magnitude and pattern of these neck surface vibration signals, our AI-based technology is currently able to identify symptoms related to airway health such as cough, throat clearing and hoarse voice with over 80% accuracy, which is important for accuracy determining severity.
Early detection of asthma and COPD flare-ups remains an unmet clinical need, but this technology may also be useful for other conditions. For example, we anticipate that this app can be extended to monitor “long COVID” as some of its symptoms – such as shortness of breath and cough – overlap with those of asthma and COPD.
With advances in wearable monitoring technology, we hope to empower and inspire patients to take charge of their airway health.