Nonlinear methods of processing physiological signals in anesthesia and vigilance: Dissertation

This thesis presents new nonlinear digital signal processing algorithms, multivariate nonlinear models, and computer based systems applied in the study of physiological signals, such as heart rate, tracheal pressure, electroencephalogram (EEG), electro-oculogram (EOG), and electromyogram (EMG), during anesthesia and reduced vigilance. Firstly, the analysis of respiration related heart rate variability, i.e. respiratory sinus arrhythmia (RSA), is studied in general. The thesis presents nonlinear FIR-median hybrid (FMH) and FIR- alpha-trimmed mean hybrid (FAH) based filters for processing respiratory sinus arrhythmia in anesthesia. The filters remove low-frequency trend and preserve the original shape of the signal without any significant distortions or attenuations. The actual phase information of RSA response is analyzed using the expirium points of tracheal pressure signal as trigger points. The averaging method used can be regarded as a special case of the large family of nonlinear methods. The different univariate indices are also compared and their limitations are considered. Secondly, the burst-suppression related changes in heart rate and DC level of EEG are characterized in anesthesia. The thesis presents a FMH based nonlinear filter to detect DC-level changes in EEG associated with bursts. The thesis also presents an averaging technique to study heart rate changes associated with EEG bursts and suppressions. The activation mechanisms related to EEG bursts and suppressions are shown with parasympathetic blockade, i.e. atropine, to mediate via the parasympathetic nervous system. However, the mechanisms related to positive pressure ventilation are not totally under parasympathetic control. Thirdly, the nonlinearity of RR interval and EEG signals in anesthesia is shown. The input- output relationship between RR interval sequence and tracheal pressure signal is modeled with Wiener modeling technique. The existence of second and third order nonlinearities is found and demonstrated. The EEG signal is modeled with nonlinear average mutual information method and linear multivariate autoregressive modeling technique. Especially, dynamic changes in correlated noise sources of the multivariate model are found during anesthesia. Finally, an objective system (called VIRTU) for vigilance studies is developed and validated in co- operation with physicians. The system is based on the analysis of EEG, EOG, and EMG signals. It uses nonlinear and linear filters for feature extraction, intelligent symbolization of extracted features compared with predetermined reference values, and rule-based decision making architec- ture.