Electrocardiogram Signal Analysis for Myocardial Infarction Detection and Reconstruction

Recent advances in biomedical signal processing have resulted in the development of many reliable medical diagnosis systems that can facilitate in providing better health-care to patients. These are accomplished with advanced techniques for accurate detection, correct classification and fast localization of disease, and efficient reconstruction of biomedical signal. Out of all the biomedical signals, electrocardiogram (ECG) is the most suitable modality for early identification of many cardiovascular diseases, as it is the easy and non-invasive way of assessment of cardiac condition. ECG signal results from the electrical activity of the myocardium cells which leads to contraction and relaxation of the heart muscles alternately in atria and ventricles. Myocardial infarction (MI) is one of the most common reason behind the mortality of human beings worldwide. Timely detection of MI in ECG signal is one of the important task in cardiac health monitoring system, which has immense importance in clinical diagnosis. On the onset of MI three basic changes are generally noticed in the leads facing the infarcted wall. ECG signals are affected by lots of noises like muscle artifact (MA), baseline wander (BW), power line interference (PLI) and white Gaussian noise (WGN). Hence tracking of all the morphological alterations in ECG recordings due to MI as well as original ECG signal reconstruction have pivotal importance in healthcare monitoring system. In this thesis, we aim to devise efficient methodologies for accurate detection of MI in ECG signal and reconstruction of the same. Fourier transform (FT) represents a signal as a summation of sinusoids and it depicts the signal in frequency domain only as there is no localization in time. Wavelet transform (WT) is an effective tool for analysis of signals with both temporal and frequency resolution levels. It overcomes the fixed window length problem of short-time Fourier transform (STFT) because of its scaling and translating parameters. Discrete WT (DWT) represents the correlation of the signal with the scaled and translated versions of the mother wavelet. The multiresolution properties of WT are explored for the extraction of the fiducial points in the ECG signal. Hence in this thesis, we have utilized wavelet transform to identify the locations affected due to abrupt changes caused by onset of MI in the frequency domain at a particular instant of time. In particular, multiscale energy analysis is carried out on the subbands resulted from WT and then using the distinct multi scale multi energy features and threshold based classification rule, the detection of MI has been performed in the ECG recordings. We have also made an attempt for the selection of suitable wavelet basis functions for detection of MI. The above proposition has limitation of providing phase information of the ECG signal. In order to overcome this bottleneck, we have employed Stockwell transform (ST) to ECG signal that provides better time resolution in high frequency and better frequency resolution in low frequency. ST uses concept of dynamic window length and also preserves the phase information of the ECG signal during signal decomposition. The phase distribution pattern as a result of MI are explored for detection of MI and to get advanced control over both time and frequency resolution, modified ST (MST) is employed for signal decomposition. The efficacy of the above proposed detection methodologies is validated using both MI and healthy ECG data. This dissertation also aims for devising reconstruction methods of ECG signal while applying to advanced cardiac health monitoring system. Interacting multiple model (IMM) has the adaptability of interchanging between several morphological representations. It offers the advantage of not necessitating user-specific parameters. This model does not necessitate a priori information about the ECG signal to initialize the filter parameters and delimitation of fiducial points of ECG signal. The particle filter (PF)-based schemes show superiority owing to their freedom from a single assumption on the signal model and noise model. Besides, it has the potential of simultaneously tracking multiple pathological and morphological changes in ECG signals due to MI. Thus, the parameters of the model are estimated by adopting the PF, so that the MI affected ECG signals can be efficiently tracked. Investigations on ECG signals signify that the IMM PF scheme can represent several MI morphologies with minimum prior information without distorting the helpful diagnostic information of ECG accurately. Better time and frequency localization ability of sparse representation approach is leveraged for efficient ECG signal reconstruction by presenting a new ECG signal reconstruction technique for efficient tracking of ECG signal in different noise driven situations. A new ECG reconstruction technique which combines IMM scheme with context aware learning based sparse representation is proposed. We adopted KSVD algorithm for dictionary learning for sparse representation of the noisy ECG signals. Different dictionaries are used to recover the ECG signals in different noise driven environments. The objective is to show the efficacy of the proposed learning based approach in tracking the multiple pathological and morphological changes occurring in ECG signals without distorting the helpful diagnostic information of ECG for better reconstruction in different noisy environments. All the proposed methods are validated with ECG data taken from PTB diagnostic ECG database and MIT-BIH arrhythmia database. Performance evaluations of these methods are compared with their related state-of-the-art methods both qualitatively and quantitatively