Investigation of breathing-disordered sleep quantification using the oxygen saturation signal

This thesis investigates the feasibility of using the non-invasive biomedical signal of oxygen saturation, or SpO2 , to diagnose a sleep disorder known as Obstructive Sleep Apnoea Hypopnoea Syndrome (OSAHS). Epidemiologically, OSAHS is the most common condition investigated by sleep clinics. In a patient suspected of having the disorder, the upper airway is obstructed during sleep and a cessation in respiration results. An apnoea is defined as a temporary cessation of breathing. Similarly, a hypopnoea is defined as any reduction in breathing (i.e., less severe than an apnoea). The work has three main objectives; the first being to establish automated evaluation procedures for methods of quantifying apnoeic activity from the SpO2 signal, the second being to accurately identify apnoeic and normal activity on a minute-by-minute basis, the third being to create a Respiratory Disturbance Index (RDI) based on the analysis which is comparable to the gold-standard Apnoea Hypopnoea Index (AHI) derived by experts. The detection of apnoeic activity is determined using three separate analyses: time domain, frequency domain, and autoregressive modelling with an incorporated amplitude criterion. A training dataset is utilised for algorithm development, and an independent dataset is employed for testing . All three methods result in comparable overall classification accuracies of: 81.2% (time domain), 82.1% (frequency domain), and 80.0% (autoregressive modelling with amplitude). In addition, particular attention is given to the resultant sensitivity, specificity, and accuracy values partitioned according to patient category; i.e., patients with OSAHS may be divided into normal, mild, moderate and severe. Lastly, a simple RDI is computed based on the automated analyses; i.e., the number of apnoeic segments detected divided by the total number of segments used. A comparison between computed RDI and AHI values for the test database show correlation values above 0.8. In conclusion, this thesis shows that through the automated analysis of the SpO2 signal, OSAHS severity in patients suspected of having the disorder can be quantified. The AR-modelling with an incorporated amplitude criterion, in particular, shows the most promise for further work in this area.