Detecção automática de eventos em sinais polissonográficos por meio de análise em tempo-frequência : aplicações à medicina do sono

Abstract

Polysomnographic signals are recorded from patients exhibiting symptoms related to sleep disorders. Obstructive sleep apnea (OSA) is one of the most widely studied sleep disorders. The assessment of the type and severity of a sleep disorder is usually based on manual identification and classification of the registered events and manual scoring of sleep stages. The usual procedure entails an overnight recording followed by the long and hard stage of manual event classification. This thesis presents a methodology to automatically detect sleep stages, in accordance to American Academy Sleep Medicine (AASM) Visual Scoring rules, using 30-second epochs taken from EEG and EMG signals. The data are processed with Empirical Mode Decomposition - EMD Method, and fed into an artificial neural network for purposed of classification. A methodology for fast screening of OSA is also provided, based on EMD decomposition of a 5-minute oronasal airway pressure signal emanating from a polysomnographic recording during the awake period, thus eschewing the need for an overnight recording. Further tests will be needed to reproduce these results and optimize the procedures, but the proposed methods seem to allow for: first, fast and reliable sleep stage scoring; second, a fast method to screen OSA patients. The presented methodology shows potential for clinical application, in order to reduce the costs and waiting times associated with the diagnosis of sleep pathologies.