ECG-Based Identification of Sudden Cardiac Death through Sparse Representations

Abstract

Sudden Cardiac Death (SCD) is an unexpected sudden death due to a loss of heart function and represents more than 50% of the deaths from cardiovascular diseases. Since cardiovascular problems change the features in the electrical signal of the heart, if significant changes are found with respect to a reference signal (healthy), then it is possible to indicate in advance a possible SCD occurrence. This work proposes SCD identification using Electrocardiogram (ECG) signals and a sparse representation technique. Moreover, the use of fixed feature ranking is avoided by considering a dictionary as a flexible set of features where each sparse representation could be seen as a dynamic feature extraction process. In this way, the involved features may differ within the dictionary's margin of similarity, which is better-suited to the large number of variations that an ECG signal contains. The experiments were carried out using the ECG signals from the MIT/BIH-SCDH and the MIT/BIH-NSR databases. The results show that it is possible to achieve a detection 30 min before the SCD event occurs, reaching an an accuracy of 95.3% under the common scheme, and 80.5% under the proposed multi-class scheme, thus being suitable for detecting a SCD episode in advance.

Keywords: ECG signals; sparse representations; sudden cardiac death.

Figure 1

Framework for ECG signal analysis: pre-processing step to obtain 1 min intervals from normal and SCD signals (yellow block), a training step for recognizing particular features from the intervals of interest (blue block), and identification of test signals through their decomposition by sparse representations (red block). In this approach, vector $$\alpha$$ is considered the feature vector.

Figure 2

Proposed multi-class scheme and its comparison with the common scheme for SCD signal classification.

Figure 3

Example of an ECG signal from a healthy subject.

Figure 4

Example of an ECG signal from a subject that suffered an SCD episode.

Figure 5

Comparison of segments from ECG signals: (a) normal, (b) pre-SCD, and (c) during SCD.

Figure 6

(a) 1-min interval from a pre-SCD signal and (b) an R-R interval extracted from it.

Figure 7

Reconstruction (synthesis) process of a signal x using its sparse representation $$\alpha$$ and a dictionary D. Coefficients in $$\alpha$$ are related to the atoms or elemental signals in D; therefore, an approximation of the original signal $$\approx x$$ can be obtained.

Figure 8

(a) Original signal, (b) signal decomposition (analysis) in elementary waveforms, (c) signal reconstruction (synthesis) by using different number of atoms, and (d) signal reconstruction with all numbers of atoms.

Figure 9

The common scheme used in SCD ECG signal classification, where an input ECG signal is identified as normal or SCD depending on its features; if a signal does not fit the characteristics of one class, then it is assumed to belong to the other.

Figure 10

Proposed multi-class scheme for SCD ECG signal classification, in which is considered that differences with respect to normal signal do not necessarily correspond to an immediate SCD but to pre-SCD intervals or even other specific causes; the number of classes (N) depends on the conditions addressed in the experiment.