Advances in Electrocardiogram Signal Processing and Analysis

William Sandham, David Hamilton, Pablo Laguna, Maurice Cohen
2007 EURASIP Journal on Advances in Signal Processing  
Since itsinvention by the Dutchman Willem Einthoven during the late 19th and early 20th centuries, when it was little more than a scientific curiosity, the electrocardiogram (ECG) has developed into one of the most important and widely used quantitative diagnostic tools in medicine. It is essential for the identification of disorders of the cardiac rhythm, extremely useful for the diagnosis and management of heart abnormalities such as myocardial infarction (heart attack), and it offers helpful
more » ... clues to the presence of generalized disorders that affect the rest of the body, such as electrolyte disturbances and drug intoxication. Recording and analysis of the ECG now involve a considerable amount of signal processing; for S/N enhancement, beat detection and delineation, automated classification, compression, hidden information extraction, and dynamic modeling. These involve a whole variety of innovative signal processing methods, including adaptive techniques, time-frequency and time-scale procedures, artificial neural networks and fuzzy logic, higher-order statistics and nonlinear schemes, fractals, hierarchical trees, Bayesian approaches, and parametric models, amongst others. This special issue reviews the current status of ECG signal processing and analysis, with particular regard to recent innovations. It reports major achievements by academic and commercial research institutions and individuals, and provides an insight into future developments within this exciting and challenging area. It is perhaps appropriate that a special issue of EURASIP JASP be devoted to ECG signal processing and analysis, since the ECG is now celebrating its centennial (Dijk and Van Loon [1]). The first paper, "Multiadaptive bionic wavelet transform: application to ECG denoising and baseline wandering reduction," by O. Sayadi and M. B. Shamsollahi, describes a new modified wavelet transform that can be used to remove a wide range of noise from an ECG signal. Signal decomposition is obtained using the bionic wavelet transform, adaptively determining both the center frequency of each scale, together with the T-function. A threshold rule is then applied. The method was tested with both real and simulated ECG signals. Results demonstrate a significantly better noise reduction compared with standard wavelet transform techniques; the average signal-to-noise ratio improved by a factor of 1.82 (best case). The method also produced better results in relation to baseline wandering calculations for both DC components and shifts. The second paper, "Hardware implementation of a modified delay-coordinate mapping-based QRS complex detection algorithm," by Matej Cvikl et al., describes a QRS detection algorithm specifically designed for hardware implementation. It uses a modified delay-coordinate mappingbased algorithm, prefiltering, and better threshold calculation methods to improve the original two-dimensional phase-space portrait. Results on the MIT-BIH arrhythmia and long-term ST databases indicate excellent sensitivity and predictivity. The next two papers cover the important topic of ECG compression. "A simple method for guaranteeing ECG quality in real-time wavelet lossy coding," by A. Alesanco and J. García, proposes a new distortion index wavelet-weighted PRD (WWPRD), which aims to provide a more realistic description of the clinical distortion of the compressed signal. The method applies the wavelet transform and the subsequent coding uses the set partitioning in hierarchical trees (SPIHT) algorithm. By thresholding the WW-PRD in the wavelet transform domain, a very precise reconstruction error was achieved, allowing clinically useful 2 EURASIP Journal on Advances in Signal Processing reconstructed signals to be obtained. Again, results from two ECG databases indicate that the method can accurately control the quality in terms of mean value with a low standard deviation. Some discussion of the impact of baseline wander and noise is included, along with a description of the clinical validation sought for these results. The second paper in this category, "Lossless compression schemes for ECG signals using neural network predictors," by K. Ramakrishnan and E. Chikkannan, presents lossless compression schemes for ECG signals based on neural network predictors and entropy encoders. Decorrelation is achieved by nonlinear prediction in the first stage and encoding of the residues is done by using lossless entropy encoders in the second stage. Different types of lossless encoders, such as Huffman, arithmetic, and runlength encoders are used. The performances of the proposed neural network predictor-based compression schemes are evaluated using standard distortion and compression efficiency measures. Selected records from the MIT-BIH arrhythmia database are used for performance evaluation. The proposed compression schemes are compared with linear predictor-based compression schemes and it is shown that about 11% improvement in compression efficiency can be achieved for neural network predictor-based schemes with the same quality and similar setup. They are also compared with other known ECG compression methods and the experimental results show that superior performances in terms of the distortion parameters of the reconstructed signals can be achieved with the proposed schemes. Segmentation and delineation is the topic covered by the fifth paper, "Combining wavelet transform and hidden Markov models for ECG segmentation," by R. V. Andreão and J. Boudy. This work aims at providing new insights into the ECG segmentation problem using wavelets. The wavelet transform has been originally combined with a hidden Markov model (HMM) framework in order to carry out beat segmentation and classification. A group of five continuous wavelet functions commonly used in ECG analysis has been implemented and compared using the same framework. All experiments were realized on the QT database, which is composed of a representative number of ambulatory recordings of several individuals and is supplied with manual labels made by a physician. A consistent set of experiments was performed, and comparable results with other published works were obtained in terms of beat delineation and premature ventricular beat (PVC) detection, independently of the type of wavelet. Optimum performance was achieved by combining two wavelet functions in the delineation stage. The next two papers address the topic of ECG modeling. In the first paper, "Modeling of electrocardiogram signals using predefined signature and envelope vector sets," by Hakan Gürkan et al., a method for modeling ECG signals is proposed based on a predefined signature and envelope vector set (PSEVS). The ECG signal is reconstructed by multiplying three parameters: predefined signature vector (PSV R ), predefined vector envelope (PEV K ), and frame-scaling coefficient (FCS). The first two measures are labeled and stored to describe the signal in the reconstruction process. An ECG signal frame is modeled as members of the sets labeled R and K and the frame-scaling coefficient in the least-mean-square sense. The method is assessed using the percentage of root-meansquare difference and also using visual inspection measures. Results show that the method provides significant data compressions ratios and low-level root-mean-square differences while preserving diagnostic information, thus significantly reducing the bandwidth required for telediagnosis. The second paper in this category, "multi-channel ECG and noise modeling: application to maternal and fetal ECG signals," by Reza Sameni et al., presents a three-dimensional dynamic model of the electrical activity of the heart. It is based on the single dipole model of the heart and is later related to the body surface potentials through a linear model which accounts for the temporal movements and rotations of the cardiac dipole, together with a realistic ECG noise model. The proposed model is also generalized to maternal and fetal ECG mixtures recorded from the abdomen of pregnant women in single and multiple pregnancies. The applicability of the model for the evaluation of signal processing algorithms is illustrated using independent component analysis. Considering the difficulties and limitations of recording long-term ECG data, especially from pregnant women, the model described in this paper may serve as an effective means of simulation and analysis of a wide range of ECGs, including adults and fetuses. The next two papers address a very useful DSP technique for ECG signals-principal component analysis (PCA). The first paper, "Principal component analysis in ECG signal processing," by Francisco Castells et al., provides a comprehensive review of the topic, and covers the fundamentals of PCA and its relationship to the Karhunen-Loève transform. Aspects on PCA related to data with temporal and spatial correlations are considered, together with adaptive estimation of principal components. Several ECG applications are reviewed, particularly where PCA techniques have been successfully employed. These include data compression, ST-T segment analysis for the detection of myocardial ischemia and abnormalities in ventricular repolarization, extraction of atrial fibrillatory waves for detailed characterization of atrial fibrillation, and analysis of body surface potential maps. The second paper, "A principal component regression approach for estimating ventricular repolarization duration variability," by Mika Tarvainen et al., introduces the idea of computing the ventricular repolarization duration (VRD) variability, not from some fiducial point (typically R-wave to Tpeak or T-end) which has a large delineation uncertainty, but from a PCA decomposition parameter. PCA decomposition is applied to T-wave segments synchronized with the R-wave. The first eigenvalue is considered to be representative of the mean T-wave shape and the second eigenvalue is of the Tvariability. Robustness is presented as the major advantage of the method, and good correlation results with classical variability measures are demonstrated. Results are presented on a stress test application. The final selection of papers deal with the analysis of the ECG for detecting specific medical conditions. The first paper, "Diurnal changes of heart rate and sympathovagal activity for temporal patterns of transient ischemic episodes William Sandham et al. 3 in 24-hour electrocardiograms," by A. Smrdel and F. Jager, presents a method for analyzing temporal patterns of transient ST segment changes compatible with ischemia, with the hypothesis that different patterns are a result of different physiological mechanisms. The study is based on 24-hour records that were divided into morning, day, and night intervals. Three temporal patterns are considered: salvo, periodic, and sporadic. Both time and frequency measures of heart rate in the neighborhood of ischemic events were analyzed using an adaptive autoregressive method with a recursive least-square algorithm, for consistent spectral tracking of heart rate, to study frequency-domain sympathovagal behavior during ischemia. Results indicate two distinct populations that differ according to mechanism and temporal patterns of ischemia. In the second paper, "Real-time cardiac arrhythmia detection using WOLA filterbank analysis of EGM signals," by Hamid Sheikhzadeh et al., novel methods of cardiac rhythm detection are proposed that are based on time-frequency analysis by a weighted overlap-add (WOLA) oversampled filterbank. Cardiac signals are obtained from intracardiac electrograms and are decomposed into the timefrequency domain and are analyzed by parallel peak detectors in selected frequency subbands. The coherence (synchrony) of the subband peaks is analyzed and employed to detect an optimal peak sequence representing the beat locations. By further analysis of the synchrony of the subband beats and the periodicity and regularity of the optimal beat, various possible cardiac events (including fibrillation, flutter, and tachycardia) are detected. Evaluation results show very good performance in clean and noisy conditions, and robustness to far-field R-wave interference. The third paper, "Corrected integral shape averaging applied to obstructive sleep apnea detection from the electrocardiogram," by S. Boudaoud et al., presents a technique called corrected integral shape averaging (CISA) for quantifying shape and shape differences in a set of signals. The method can be used to account for signal differences which are purely due to affine time warping (jitter and dilation/compression), and hence provides access to intrinsic shape fluctuations. CISA can also be used to define a distance between shapes which have useful mathematical properties, and the procedure also allows joint estimation of the affine time parameters. Numerical simulations are presented to support the algorithm. CISA provides a well-defined shape distance, which can be used in shape clustering applications based on distance measures such as k-means. An application is presented in which CISA shape clustering is applied to P-waves extracted from the ECGs of subjects suffering from sleep apnea. The resulting shape clustering distinguishes ECG segments recorded during apnea from those recorded during normal breathing with a sensitivity of 81% and specificity of 84%. The fourth paper, "Time-frequency analysis of heart rate variability for neonatal seizure detection," by M. B. Malarvili et al., proposes a technique for detecting epileptic seizures from the ECG in the human neonate. The suitability of heart rate variability (HRV) as a tool for seizure detection in newborns is explored. Features of HRV in different frequency bands have been obtained by means of quadratic time-frequency dis-tributions (TFDs). The first conditional moment of HRV, which is the mean/central frequency in the LF band (0.03-0.07 Hz) and the variance in the HF band (0.15-0.6 Hz) is shown to have the ability to discriminate the newborn seizure from nonseizure. In the final paper in this category, "Clustering and symbolic analysis of cardiovascular signals: discovery and visualization of medically relevant patterns in long-term data with limited prior knowledge," by Zeeshan Syed et al., automated techniques are presented for analyzing large amounts of cardiovascular data without the requirement of a priori knowledge of disease states. The process begins by transforming continuous waveform signals into symbolic strings derived from the data. Morphological features are used to partition heartbeats into clusters by maximizing the dynamic time-warped sequence-aligned separation of clusters. Each cluster is then assigned a symbol, and the original signal is replaced by the corresponding sequence of symbols, thus greatly reducing the amount of data. The sequence analysis is used to discover rhythms, transient patterns, abnormal changes in entropy, and clinically significant relationships among multiple streams of physiological data. The method was tested on cardiologist-annotated ECG data for 48 patients. The labeling process for heartbeats showed 98.6% agreement with the assessment of the cardiologist, and often provided finer-grain distinctions. Using no prior knowledge, cases with a number of different arrhythmias were successfully detected. ACKNOWLEDGMENTS
doi:10.1155/2007/69169 fatcat:rxxotykyh5dw5kfy3rvtrh67hq