Automatic Classification of Polysomnographic Respiration Signals

Within the project on biomedical signal analyses with artificial neural networks, recent research is focused on polysomnographic signal analyses. In polysomnographics (PSG) a large variety of physiological signals, obtained during a patients sleep, are analyzed and classified. This research focuses on the detection of obstructiveand central-apnea (absence of air flow through the nose), paradoxical respiration, normal respiration and artifacts in the noseflow , diaphragm and thorax signals. The signals are analyzed and characteristic features are extracted by different methods: FFT, correlation, filtering. The classification is implemented by the use of classical Bayes classification combined with relative modern neural network approaches. Multi-layer perceptrons (MLP) and Kohonen networks are used for probability density estimation of the features of the different classes. Multi-level classification and multi-level feature extraction were necessary. A dynamically, interactive, evolving..