| This dissertation presents computer algorithms for the automated detection and classification of sleep disordered breathing events from conventional polysomnography (PSG) data in an undiagnosed working population. The algorithms are based on the analysis of the respiratory inductive plethysmography (RIP) signal to recognize apneic and hypopneic patterns. An apnea is defined as a period of no RIP movement lasting more than 10 s. Hypopnea is a pattern of sum RIP reduction by at least 20% from a single-breath baseline lasting more than two breaths followed by a return to within 90% of baseline. Both apnea and hypopnea definitions require an associated desaturation of at least 2%. When validated against manual scoring of 10 PSG records, the algorithm had a 99.1% detection sensitivity for desaturation, 73.6% for apnea, 84.1% for hypopnea and 93.1% for apnea plus hypopnea.; To investigate the physiological significance of computer-detected hypopneas, especially those with mild desaturation, we measured cardiovascular and EEG responses during and following each SDB event. During NREM sleep in 8 subjects, hypopneas with less than 4% desaturation were associated with significant average blood pressure (BP) and heart rate (HR) increases towards their resolution with 23.4% of them resulting in EEG arousals (2 to 15-s increases in EEG frequency). Furthermore, hypopneas associated with arousals resulted in significantly higher BP and HR responses than non-arousing hypopneas at all desaturation levels. We incorporated measurements of HR, BP, desaturation and arousal into a fuzzy system classifier to assign significance values to computer-detected hypopneas based on the magnitudes of the responses near the termination of the event. 36% of all hypopneas had a significance value greater than the midrange value and were determined by the system to be significant. Automated detection and analysis of the acute responses of events that are usually overlooked by conventional manual scoring can be very important for the assessment of sleep in population studies. |
