SVM classification of CWT signal features for predicting sudden cardiac death

Abstract

Sudden Cardiac Death (SCD) is a major health problem that is responsible for most of all the heart disease deaths. The Ventricular Tachyarrhythmia's (VT's), especially the Ventricular Fibrillation (VF) are the primary cause of the SCD's. This paper presents a classification method using Support Vector Machine (SVM) algorithm for predicting if there is an SCD occurrence in a signal. This is carried out by comparing certain characteristic features of the ECG signal of a normal healthy person with that of the unhealthy patient prone to SCD. In the time domain, the ECG signal has to be monitored continuously for long hours, which is not feasible and moreover the cardiac arrest in SCD cases occurs for a very short time which is preceded and followed by normal ECG. The noise and some electrical disturbances during measurement also affect the signal feature measurements. So, the signal is transformed into another domain using Wavelet Transformation method (Continuous Wavelet Transform (CWT) to be precise) to extract certain features of the signal and study their pattern while comparing the abnormal ECG signal with that of a normally running ECG signal. The main features that were extracted are the R-peaks, R to R intervals, QRS complexes, QRS complex durations, T-Wave durations and the QT intervals. CWT was used to extract the features information from the ECG signals providing the base. The tests were performed on the data signals taken from the Physionet database [12]. General reminders and related works2.1. Sudden cardiac death Sudden Cardiac Death (SCD) [2] is an unexpected death of a person caused due to Sudden Cardiac Arrest (SCA). SCD and SCA are sometimes used interchangeably. There is a sudden loss of cardiac function and the heart abruptly stops beating. It can also be said as the RECEIVED

Time adaptive ECG driven cardiovascular disease detector

“…Lastly, sudden cardiac death has also been investigated in several studies, that are different in research contexts, research objectives, datasets, used variables and validation procedures, making the comparison of studies challenging with our model. The most frequently observed research objective was the classification of SCD and non-SCD via traditional machine learning [54] and by using Heart Rate Variability features as predictors [55,56]. One recent study, Kaspal et al [57], used a CNN combined with a Recurrence Complex Network (RCN) to enhance the accuracy of SCD classification.…”

ECG-based prediction algorithm for imminent malignant ventricular arrhythmias using decision tree

“…The uncertainty is due to different possible prediction times before the onset of imminent VT/VF that are demonstrated in the past studies. The time would be one hour, several minutes or even a few seconds preceding the onset [15,13,8,14]. If possible prediction time is short, self-rescue time of potential patients is also short.…”

“…The dependence is reduced when the number of features is six and above. The recommendation is according to previous studies that performed prediction using different numbers of features, which was either two or greater than five [15,13,2,8].…”

“…Numerous features derived from intervals, amplitudes and waveform morphology of ECG components are useful to detect underlying VT/VF. For example, mean of RR intervals (mRR), QRS duration (QRSd), mean of QT intervals, and T-wave alternans could be associated with increased risk of MVA [3,4,5,6,7,8]. The features may prolong, reduce or precede VT/VF.…”

“…Second, the time required for an algorithm to execute a prediction for imminent MVA is also unknown, because it is usually neglected in existing studies [15,13,7,2,8]. Instead of https://doi.org/10.1371/journal.pone.0231635.g001…”