An Approach to ECG Delineation using Wavelet Analysis and Hidden Markov Models

Analysis and interpretation of the human electrocardiogram (ECG) is seriously impaired by high levels of noise and a wide variety of signal morphologies. These problems relate to the cardiac state of the patient and nature of the sensing process. A useful property, specific to the ECG signal in contrast to for instance EEG signals, is its evident cyclical nature. A typical ECG follows a fixed sequence of distinctive and well-identifiable components, called the P-, QRS-, and T-wave. Using and exploiting this property, it was possible to develop an integrated approach to ECG analysis, combining multi-resolution Wavelet Analysis (WA) for signal analysis with Hidden Markov Models (HMM) for tracking the typical ECG cycle. Furthermore, the HMM allows for modeling the statistical properties of an ECG. The state durations implicit in a standard Markov model are ill-suited to model those of the ECG features, therefore the use of Hidden Semi-Markov Models (HSMM) was also investigated. The combination of these two techniques was examined in three different architectures; (1) the WA of the ECG as input for the HMM, (2) the WA of the ECG as input for the HSMM, and (3) an approach called the WTSign method. In the WTSign method, edges in the ECG are localized, and features extracted from the edges serve as input for the HMM. These approaches were tested and evaluated on the manually annotated database (MIT-BIH QT-database), which is regarded as a very important benchmark for ECG analysis. The HMM obtained a sensitivity Se = 99, 11% for QRS detection and Se = 66, 86% for T-wave detection. While for the HSMM, sensitivities of Se = 98, 79% and Se = 83, 49% were achieved for QRS and T-wave detection respectively. Finally, the WTSign method obtained a sensitivity Se = 99, 40% for QRS detection and a sensitivity Se = 94, 65% for T-wave detection.