Obstructive Sleep Apnoea (OSA) is a sleep related disorder, which causes repetitive collapse of upper airways during sleep leading to serious consequences. Aging populations and rising number of obese people are making OSA an increasingly important public health problem. At present, over 90% of the patients remain undiagnosed. Gold standard of OSA diagnosis, Polysomnography (PSG), requires the patient to spend full night at the sleep clinic attached to a multitude of measurement leads. Therefore PSG is inconvenient and expensive, making it unsuited for community screening.
Snoring, which is a common symptom of OSA, carries vital information about the status of the upper airways, though it is rarely being used for diagnosis. Several advantages of using snoring sound are non-invasive analysis, cost effectiveness, possibility of automated scoring and unattended study. This can be developed as a single measurement, which is potentially superior to PSG and other portable monitoring devices for screening OSA. In this research project we develop a clinical database of full night snoring sounds. We propose algorithms to segment and categorise recorded snoring signal into silence and voiced/unvoiced snoring sounds. These algorithms allow us to extract all the snores from the full night recording of sounds around 99% accuracy. Then we analyse the extracted snores from the subjects to track the development of apnoea events starting from the time a subject goes to sleep, using pitch-jitter information for mild OSA patients.
We introduce a novel feature, called intra snore pitch jumps (ISPJ), in a view to diagnose OSA. We tested the feature using a clinical database of full night snoring from 29 subjects. Sensitivity values of 83-1OO% (at specificity of 50-80%) were obtained at several severity levels of OSA. Achieved sensitivity levels indicate that feature ISPJ can be used for screening OSA at home setting using a simple monitor.