Prompt prediction of successful defibrillation from 1-s ventricular fibrillation waveform in patients with out-of-hospital sudden cardiac arrest

Abstracts

Purpose

Ventricular fibrillation (VF) is a common cardiac arrest rhythm that can be terminated by electrical defibrillation. During cardiopulmonary resuscitation, there is a strong need for a prompt and reliable predictor of successful defibrillation because myocardial damage can result from repeated futile defibrillation attempts. Continuous wavelet transform (CWT) provides excellent time and frequency resolution of signals. The purpose of this study was to evaluate whether features based on CWT could predict successful defibrillation.

Methods

VF electrocardiogram (ECG) waveforms stored in ambulance-located defibrillators were collected. Predefibrillation waveforms were divided into 1.0- or 5.12-s VF waveforms. Indices in frequency domain or nonlinear analysis were calculated on the 5.12-s waveform. Simultaneously, CWT was performed on the 1.0-s waveform, and total low-band (1–3 Hz), mid-band (3–10 Hz), and high-band (10–32 Hz) energy were calculated.

Results

In 152 patients with out-of-hospital cardiac arrest, a total of 233 ECG predefibrillation recordings, consisting of 164 unsuccessful and 69 successful episodes, were analyzed. Indices of frequency domain analysis (peak frequency, centroid frequency, and amplitude spectral area), nonlinear analysis (approximate entropy and Hurst exponent, detrended fluctuation analysis), and CWT analysis (mid-band and high-band energy) were significantly different between unsuccessful and successful episodes (P < 0.01 for all). However, logistic regression analysis showed that centroid frequency and total mid-band energy were effective predictors (P < 0.01 for both).

Conclusions

Energy spectrum analysis based on CWT as short as a 1.0-s VF ECG waveform enables prompt and reliable prediction of successful defibrillation.

Appendix

Continuous wavelet transform (CWT)

CWT is defined as the convolution of a signal and an analyzing wavelet (ψ).

$$ C(\alpha ,\tau ) = \sum\limits_{t = 0}^{n - 1} {{\text{signal}}\,(t)\psi *\left( {{\frac{t - \tau }{\alpha }}} \right)} $$

where the asterisk indicates a complex conjugate, α indicates the scale (dilation), τ indicates a time shift, and signal (t) denotes sample t in the VF waveform segment of length n. In this study, a complex Morlet wavelet was employed as the analyzing wavelet (ψ) [16, 17].

The proportional contribution to the signal energy at a specific scale and a location τ is given by the two-dimensional wavelet energy density function:

$$ E(\alpha ,\tau ) = {\frac{{[C(\alpha ,\tau )]^{2} }}{{C_{\text{g}} }}} $$

where C _g is the wavelet-dependent admissibility constant that ensures conservation of energy in wavelet space [30].

Frequency-domain analysis

Fourier transform is defined as follows:

$$ F({\text{fi}}) = \sum\limits_{t = 0}^{n - 1} {{\text{signal}}\,(t)e^{{ - j2\pi {\text{fi}}n}} } $$

where fi indicates frequency and signal (t) denotes sample t in VF waveform segment of length n.

Power spectral density (PSD) describes how the power of a signal is distributed with frequency. Mathematically, it is defined as the squared modulus of the Fourier transform of the signals.

$$ {\text{PSD}}({\text{fi}}) = [F({\text{fi}})]^{2} $$

$$ {\text{Peak}}\,{\text{power}}\,{\text{frequency}}\,({\text{Hz}}) = \arg\max \,{\text{PSD}}({\text{fi}}) $$

where fi indicates frequency and argmax indicates the fi attaining a max value of PSD.

$$ {\text{Centroid}}\,{\text{frequency}}\,({\text{Hz}}) = {\frac{{\sum {\text{fi}} \, \times \,{\text{PSD(fi)}}}}{{\sum {\text{PSD(fi)}} }}} $$

$$ {\text{Amplitude}}\,{\text{spectral}}\;{\text{are}}\,({\text{AMSA}})\,({\text{mVHz}}) = \sum {{\text{fi}}\, \times \,{\text{ASD(fi)}}} $$

where ASD(fi) indicates amplitude spectral density, calculated as

$$ {\text{ASD(fi)}} = \sqrt{{{\text{PSD(fi)}}\, \times \,{\frac{\text{fi}}{2}}}} $$