A unified procedure for detecting, quantifying, and validating electrocardiogram T-wave alternans

Non-Standard Electrode Placement Strategies for ECG Signal Acquisition

BACKGROUND

Wearable technologies for monitoring cardiovascular parameters, including electrocardiography (ECG) and impedance cardiography (ICG), propose a challenging research subject. The expectancy for wearable devices to be unobtrusive and miniaturized sets a goal to develop smarter devices and better methods for signal acquisition, processing, and decision-making.

METHODS

In this work, non-standard electrode placement configurations (EPC) on the thoracic area and single arm were experimented for ECG signal acquisition. The locations were selected for joint acquisition of ECG and ICG, targeted to suitability for integrating into wearable devices. The methodology for comparing the detected signals of ECG was developed, presented, and applied to determine the R, S, and T waves and RR interval. An algorithm was proposed to distinguish the R waves in the case of large T waves.

RESULTS

Results show the feasibility of using non-standard EPCs, manifesting in recognizable signal waveforms with reasonable quality for post-processing. A considerably lower median sensitivity of R wave was verified (27.3%) compared with T wave (49%) and S wave (44.9%) throughout the used data. The proposed algorithm for distinguishing R wave from large T wave shows satisfactory results.

CONCLUSIONS

The most suitable non-standard locations for ECG monitoring in conjunction with ICG were determined and proposed.

Improved robust T-wave alternans detectors

New statistical and spectral detectors, the modified matched pairs t test, the extended spectral method and the modified spectral method, were proposed for T-wave alternans (TWA) detection gaining robustness according to trend and single-frequency interferences. They were compared to classic detectors such as matched pairs t test, unpaired t test, spectral method, generalized likelihood ratio test and estimated TWA amplitude within a simulation framework and applied to real data. The optimal detection threshold was selected by using a full Monte-Carlo simulation where signals, with and without alternans episodes, were corrupted by Gaussian noise with different power and single-frequency interferences with different tones. All the combinations of noise and frequency were selected and repeated 500 times in order to compute probability of detection ([Formula: see text]) and the false alarm probability ([Formula: see text]), providing ROC curves. The study group consisted of 50 patients with implantable cardioverter-defibrillator (age: [Formula: see text]; LVEF: [Formula: see text]), who were paced (ventricular pacing) at 100 bpm. Two-minute recordings were analyzed. The XYZ orthogonal lead system was used. The best performance was reached by using the modified matched pairs t test (in comparison with the spectral method and other reference methods).

Nonparametric signal processing validation in T-wave alternans detection and estimation

Although a number of methods have been proposed for T-Wave Alternans (TWA) detection and estimation, their performance strongly depends on their signal processing stages and on their free parameters tuning. The dependence of the system quality with respect to the main signal processing stages in TWA algorithms has not yet been studied. This study seeks to optimize the final performance of the system by successive comparisons of pairs of TWA analysis systems, with one single processing difference between them. For this purpose, a set of decision statistics are proposed to evaluate the performance, and a nonparametric hypothesis test (from Bootstrap resampling) is used to make systematic decisions. Both the temporal method (TM) and the spectral method (SM) are analyzed in this study. The experiments were carried out in two datasets: first, in semisynthetic signals with artificial alternant waves and added noise; second, in two public Holter databases with different documented risk of sudden cardiac death. For semisynthetic signals (SNR = 15 dB), after the optimization procedure, a reduction of 34.0% (TM) and 5.2% (SM) of the power of TWA amplitude estimation errors was achieved, and the power of error probability was reduced by 74.7% (SM). For Holter databases, appropriate tuning of several processing blocks, led to a larger intergroup separation between the two populations for TWA amplitude estimation. Our proposal can be used as a systematic procedure for signal processing block optimization in TWA algorithmic implementations.

Real-time electrocardiogram P-QRS-T detection-delineation algorithm based on quality-supported analysis of characteristic templates

The main objective of this study is to introduce a simple, low-latency, and accurate algorithm for real-time detection of P-QRS-T waves in the electrocardiogram (ECG) signal. In the proposed method, real-time signal preprocessing, which includes high frequency noise filtering and baseline wander reduction, is performed by applying discrete wavelet transform (DWT). A method based on signal first-order derivative and adaptive threshold adjustment is employed for real-time detection of the QRS complex. Moreover, detection and delineation of P- and T-waves are achieved by correlation analysis conducted between signal and their templates. Besides, signal quality is investigated online, and if the quality of the analysis window is unacceptable, then the algorithm will guess (estimate) the locations of P- and T-waves. The operating characteristics of the proposed algorithm are evaluated by its implementation to an artificially generated ECG signal whose quality is adjustable from the best (Quality, 100%) to the worst (Quality, ≤40%) cases based on the random-walk noise theory. The algorithm was applied to the MIT-BIH arrhythmia database, QT database, and Physionet/CinC challenge 2011competition database. The obtained results, which were based on the QT database, showed sensitivity and positive predictivity of Se=99.63% and P+=99.83%, Se=99.83% and P+=99.98%, and Se=99.74% and P+=99.89% for the detection of P-, QRS-, and T-waves, respectively, and the obtained results, which were based on the MIT-BIH arrhythmia database, showed Se=99.81% and P+=99.70% for the detection of the QRS complex. Moreover, it will be shown that the results of the proposed method are reliable for a minimum signal quality value of 70%. According to numerical assessments, 8-ms after the occurrence of R-wave, its location will be identified by the computer code of the proposed algorithm. This parameter is 198-ms and 177-ms for P- and T-waves, respectively.

Electrocardiogram signal quality assessment using an artificially reconstructed target lead

In real applications, even the most accurate electrocardiogram (ECG) analysis algorithm, based on research databases, might breakdown completely if a quality measurement technique is not applied precisely before the analysis. The major concentration of this study is to describe and develop a reliable ECG signal quality assessment technique. The proposed algorithm includes three major stages: preprocessing, energy-concavity index (ECI) analysis and a correlation-based examination subroutine. The preprocessing step includes the removal of baseline wanders and high-frequency disturbances. The quality measurement based on ECI includes two separate stages according to the energy and concavity of the ECG signal. The correlation-based quality measurement step is mainly established by using the correlation between ECG leads estimated by applying a suitably trained neural network. The operating characteristics of the proposed ECI are sensitivity (Se) of 77.04% with a positive predictive value (PPV) of 90.53% for detecting high-energy noise. The correlation-based technique achieved the best scores (Se = 100%; PPV = 98.92%) for detecting high-energy noise and for recognising any other kind of disturbances (Se = 92.36%; PPV = 94.77%). Although ECI analysis acts effectively against high-energy disturbances, very poor performance is obtained in cases where the energy of the disturbances is not considerable. However, the correlation-based method is able to find all kinds of disturbances. For officially evaluating the proposed algorithm, an entry was sent to the Computing-in-Cardiology Challenge 2011 on 27 February 2012; a final score (accuracy) of 93.60% was achieved.