Non-intrusive medical-grade accuracy respiratory monitoring is one of the two unresolved challenges in design of wearable vital signs monitoring system, besides blood pressure monitoring. The challenge of respiratory monitoring includes automatic breath detection, respiratory phase classification, and extraction of other respiratory related parameters. The main problem is that the sensed respiratory physiological signals could be heavily corrupted by artefacts and interference from non-relevant sources such as movement and other unwanted physiological signals collected at the same time. This thesis proposes a novel system which can detect respiratory activity from tracheal sounds recorded by a miniature wearable sensor with limited power budget, and can be tailored for variety of clinical scenarios by combining novel state-of-the-art techniques at both hardware and software level.
The respiratory monitoring system presented in this thesis consists of two parts: one is a hardware platform for sensing breath sounds and the other is an algorithm for respiratory activity analysis. The low power sensor platform has been optimised for breath sound data acquisition by a combination of: a custom engineered acoustic chamber and a MEMS microphone; and optional power efficient data compression techniques which include the use of novel fixed length adaptive sample size (FLASS) frames optimised for Bluetooth® Smart wireless transmission protocol. The respiratory activity algorithm is based on 3-stage backward adaptive approach, using previous detected respiration’s parameters to fine tune the system and then balancing the system’s sensitivity and specificity in various ways to suit the need of different applications. These include sleep apnoea diagnosis, sudden unexpected death prevention in both epilepsy and babies, early warning scoring in hospitals and management of chronic conditions such as chronic obstructive pulmonary disease and asthma.
A clinical trial focused on sleep apnoea detection was conducted in the National Hospital for Neurology and Neurosurgery, London, UK, in order to verify the respiratory monitoring system. Questionnaire results show that all participants scored the novel wearable respiratory sensor in the range 4 to 5, where 5 represents the best possible scoring in a number of attributes related to wearability. Experimental results show the proposed automatic apnoea and hypopnoea detection algorithm has 88.6% sensitivity and 99.6% specificity when comparing its performance with a specialist clinician considered as a gold standard. In comparison a state-of-the-art automatic ambulatory sleep diagnosis system has 14.3% sensitivity and 99.3% specificity.