Detecção e segmentação automática de batimentos cardíacos do eletrocardiograma por modelagem matemática e combinação das transformadas Wavelet e de Hilbert

• Ano de defesa: 2013
• Autor(a) principal: Madeiro, João Paulo do Vale
• Orientador(a): Cortez, Paulo César
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Teleinformática; Eletrocardiografia; Wavelet, transformada
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/5655

A utomatic diagnostic aid systems aim the extraction of speci c parameters in order to support the analysis of a patient's physiological conditions possibly using computing algorithms. In the context of cardiology, such systems are particularly important when applied over long-term ECG signals, for example the 24-h holter examinations. The digital signal processing techniques for ECG waves segmentation and automatic feature extraction, which are proposed in this thesis, cover various research elds. Firstly, the proposed system performs QRS complex detection and segmentation, which is related to ventricular depolarization. The used methodology combines the adaptive threshold technique, Hilbert and Wavelet transforms and the rst-derivative lter with a new approach of preprocessing suppression over the whole ECG signal and selection of Wavelet scale factor for a given predominant QRS morphology. As output information we obtain the RR time-series (tachogram), the time-series of QRS complex durations and amplitudes. In the second stage, the developed system performs T-wave detection and segmentation, whose waveform is related to ventricular repolarization activity. It is proposed a new mathematical model concerning the possible T-wave morphologies based on a Gaussian function, modi ed by a mathematical procedure to insert asymmetry. Once the template is computed, cross-correlation and an approach based on the computation of Trapezium's area are used to locate, respectively, the peak, or the peaks for biphasic waves, and end point of each T-wave throughout the whole raw ECG signal. Among the metrics derived from the detected ducial points, we emphasize the QT intervals, which are the time intervals between the QRS onset and the T-wave end. After the segmentation of the ECG waves, we perform two important case studies using the ducial points and segments detected in the previous stages: ventricular activity subtraction in intracardiac atrial brillation electrogram and heart hate variability (HRV) analysis for a set of elderly patients which were selected in the Geriatric Outpatient Clinic of the Walter Cantidio University Hospital. After evaluating the overall methodology of QRS detection and segmentation over various manually annotated databases, inclusive the public MIT-BIH Arrhythmia database and QT database, we have obtained the following detection rates and delineation time errors: sensitivity of 99.51%, positive predictivity of 99.44%, QRS onset time error of 2.85 9.90 ms and QRS o set time error of 2.83 12.26 ms. Regarding T-wave detection and delineation, the proposed method has attained sensitivity of 99.48%, positive predictivity of 99.53%, and average time errors of 0.51 8.06 ms, for T-wave peak location, and 0.11 11.73 ms, for T-wave end location. Regarding the rst case study concerning the use of the ducial points detected from the segmented QRS complexes and T-waves over intracardiac atrial brillation electrogram, the method of ventricular activity subtraction has attained a signi cant attenuation for frequencies above 10 Hz, and also for components of frequency range around 3 Hz to 6 Hz, respectively due to ventricular depolarization and repolarization subtraction. For the second application, the analysis of the evolution of heart rate variability metrics in frequency domain associated to sympathetic branch activity allows recognizing speci c tendencies regarding aspects of proper functioning/dysautonomia of the autonomic nervous system for each predetermined elderly class according to the concepts of frailty phenotype: frail, pre-frail and robust ones. The overall results suggest that the set of methodologies developed for ECG waves segmentation provides high rates of accurate and robust detections for a wide variety of morphologies, such that they can be applied in various situations for aid to diagnosis. Given the set of possible metrics and time-series which can be extracted from the ECG signals, after their segmentation, the referred methods can support projects of clinical research and development of markers/indicators of adverse cardiovascular events.