Investigation of Real-Time Magnitude Squared Coherence for Discrimination of Arrhythmias from Intracardiac Electrograms

Proper functioning of implantable anti-arrhythmia devices requires, in part, accurate detection of various cardiac arrhythmias. Both the morphology and amplitude of the intracardiac electrogram signal significantly affect the accuracy of current arrhythmia discrimination schemes in discriminating fibrillatory cardiac rhythms (FIB) from non-fibrillatory cardiac rhythms (NON-FIB). Poor detection schemes result in unnecessary administration of electrical shocks during nonpathologic rhythms or the absence of shocks during fibrillatory rhythms. Previous studies show that Magnitude-Squared Coherence (MSC) is a reliable algorithm for detecting fibrillation. However, use of MSC in an implantable device will require that it be performed in real-time and that it discriminates FIB from NON-FIB with better sensitivity and specificity than rate or PDF. Three real-time Fourier Transform algorithms (regular Fast Fourier Transform (FFT), Moving Discrete Fourier Transform (MDFT) and Moving Fast Fourier Transform (MFFT) were compared in computational intensity and memory storage. Real-time MSC estimates were computed and compared for variable frequency bands and for variable window lengths for examples of both fibrillatory and nonfibrillatory electrogram recordings. In addition, real-time MSC, Rate and PDF were compared in their ability to detect onset of fibrillation using a simulated database of which demonstrate transition from non-fibrillation and fibrillation. Results indicate that for a single data point Fourier Transform update, the MDFT requires the least computation and the MFFY requires the most memory storage. Both MFFT and MDFT require less than half of the computations as that of the regular FFT. For a q-data point Fourier Transform update, as q increases, the computational intensity for the MFFT and the MDFT increase while the memory required for storage for the MFFT decreases. As the overlap between adjacent Fourier Transform estimates reaches 50%, there is no difference between the MFFT and the FFT in computational intensity or memory storage. The MDFT algorithm results in significant saving in MSC computation if the Fourier Transform need only be calculated at few frequencies. The recursive error of the MDFT is minimal and does not affect the ability of the MDFT MSC estimate to discriminate FIB from NON-FIB rhythms. Results suggest that only a narrow band of frequencies are necessary to discriminate fibrillatory from non-fibrillatory rhythms. Our result suggest that MMSC (Mean Magnitude Squared Coherence) in the 20-50 Hz frequency band best differentiate non-ventricular fibrillation rhythms (NON-VF) and ventricular fibrillation rhythms (VF) for the electrophysiologic testing (EP) lead configuration and MMSC in the 5-10 Hz frequency band best differentiate NON-VF and VF for the automatic implantable cardioverter/defibrillator (AICD) lead configuration. For the AICD lead configuration, the ability of MMSC to differentiate NON-VF from VF in 0-60 and 20-50 Hz bands was less significant. Although the rate algorithm for the EP leads exhibited a total accuracy over 89% for atrial rhythms and total accuracy over 90% for ventricular rhythms, MMSC performed better (total accuracy I00% for the EP atrial rhythms and total accuracy 96.3% for the ventricular rhythms). This research showed that the rate algorithm failed to differentiate VF and NON-VF rhythms for the AICD lead configuration, which is used in current commercial devices. In addition, this research showed that rate algorithm provided a better differentiation of FIB and NON-FIB in the right ventricle apex (RVA) lead configuration than right ventricle outflow tract (RVOT) lead configuration for the EP lead configuration