Acoustic wearable systems and data for the monitoring of respiratory health

Lead Research Organisation: University of Cambridge
Department Name: Computer Laboratory

Abstract

Respiratory conditions are the third most prevalent cause of deaths worldwide. They are typically assessed by listening to lung sounds using a stethoscope. Using these sounds, respiratory diseases can be diagnosed by identifying the presence of abnormal sounds, including crackles and wheezes. However, this diagnosis is often prone to human error due to the limitations in human hearing. The diagnoses can also vary between different physicians, leading to inconsistencies. It is also difficult to continuously monitor respiratory sounds using a stethoscope, making it challenging to assess disease progression. Therefore, it is necessary to develop objective methods of diagnosing and monitoring abnormal respiratory sounds.

To achieve this, this project aims to develop a wearable sensor for continuous and automatic detection of respiratory abnormalities. Signal processing techniques (such as mel-frequency cepstral coefficients) and machine learning models will be developed to analyse respiratory sounds. This analysis will take the form of differentiating between normal and abnormal lung sounds and classifying the types of abnormalities. Open-source respiratory sound datasets will be used to train and test these models. Algorithms will also be developed to isolate respiratory sounds from external noise sources, including talking and other ambient sounds. Finally, a sensor-based system will be designed which is capable of recording respiratory sounds without impeding on a user's comfort and normal activities. Different acoustic sensors will be compared and the most effective developed into the wearable sensor system. Once the sensor system has been designed, firmware will be written to process the signals on board, allowing for continuous diagnosis and monitoring of respiratory conditions.

The project is aligned to a number of EPSRC research areas, with the most prominent being "Clinical technologies", "Artificial intelligence technologies", and "Digital signal processing".

Publications

10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/S023046/1 01/10/2019 31/03/2028
2258809 Studentship EP/S023046/1 01/10/2019 30/09/2023 Kayla-Jade Butkow