Detecção das ondas R de sinais de eletrocardiograma por meio da Local Mean Decomposition e da Empirical Mode Decomposition

Abstract

The analysis of the Electrocardiogram (ECG) signal is fundamental for assessment of cardiac electrical activity in individuals, being important for the identification of arrhythmias and heart diseases. This analysis involves the identification of characteristic waves (P, Q, R, S, and T), as well as their beginning and end. The identification of the R wave, specially, is crucial for calculating heart rate and its variability. Thus, this work aimed at detecting R waves using Local Mean Decomposition (LMD), Empirical Mode Decomposition (EMD), and the Pan-Tompkins Method (PTM) applied to four ECG signal databases: MIT-BIH Normal Sinus Rhythm Database (healthy individuals), Congestive Heart Failure RR Interval Database (individuals with heart failure), MIT-BIH Supraventricular Arrhythmia Database (individuals with supraventricular arrhythmia), and CU Ventricular Tachyarrhythmia Database (individuals with ventricular tachyarrhythmia). The correct or incorrect identification of R waves is verified based on the annotation of their occurrence made by a specialist. For each technique, the average Execution Time (TExec) was calculated. PTM was the fastest, followed by EMD, and lastly, LMD. However, LMD presented an average TExec ranging from 1 to 5 seconds for processing 8 or 10-minute signals; thus, even this technique could be applied for online processing. Additionally, True Positives (TP), False Positives (FP), and False Negatives (FN) were calculated for each technique, and from these values, Sensitivity (Se), Positive Predictive Value (PPV), and Error Rate (Error%) were obtained. For the normal sinus rhythm and heart failure databases, all three techniques showed excellent performance with very high Se and PPV and very low Error%. For ventricular tachyarrhythmia signals, all techniques showed poor performance, with LMD having the lowest error rate (around 30%). For the supraventricular arrhythmia database, PTM showed good performance, followed by LMD and EMD (with signals reconstructed from the Intrinsic Mode Functions IMF1-IMF2 and IMF1-IMF3) with intermediate performances. For EMD (reconstructed with IMF1), the error rate was unacceptable (>40%), as was the very low Sensitivity (around 60%). The results indicated the classical PTM technique as the one with the best overall performance. However, the results of applying LMD and EMD can be considered promising, as the performance of these techniques is highly dependent on the detection criterion used. In this work, amplitude threshold-based criteria were used for both LMD and EMD; however, other criteria described in the literature deserve investigation.