Intracardiac electrogram transformation. Morphometric implications for implantable devices.

Over 75,000 antitachycardia devices (ATDs) have been implanted since initial Food and Drug Adrninistra-tion approval in 1985 and have yielded dramatic survival rates. These devices, although life-saving, rely principally on simple measures of the heart rate for arrhythmia detection. The ATDs are highly sensitive, but their low specificity of diagnosis results in episodes of inappropriate therapy ranging as high as 10 to 41 % of all shocks delivered (1-5). Morphologically based algorithms have been demonstrated to improve levels of specificity dramatically while maintaining high levels of sensitivity (6-1 1); however, these algorithms tend to be computationally complex, placing unacceptably large demands on battery power (12,13). Since ATDs are small battery-operated implants, small algorithmic demands on battery power are essential for device longevity. Correlation waveform analysis (CWA), a morphologically based algorithm, has emerged as a promising technique for more specific assessment of intracardiac elec-trogram (IEGM) rhythms. Correlation waveform analysis lntracardiac Electrocardiogram Transformation Morris et al. 125 has previo~~sly been demonstrated to separate mono-A) morphic ventricular tachycardia (MVT) and ventricular fibrillation (VF) from sinus rhythm (SR) and performs as well, if not better, than other morphologically based algorithms when classifying MVT from SR (12). However , the power demand of CWA has led to its exclusion from ATD algorithm implementation. If the number of data points required for accurate analysis by CWA were sufficiently reduced, the diminished power demand con-comitant with such a reduction could make CWA an attractive discriminant. The Karhunen-Loeve transformation and feature selection of the IEGM was examined in this study to determine whether the Karhunen-Loeve transformation with feature selection yields increased specificity and reduced the power demand of morphologically based algorithms such as CWA. The Karhunen-Loeve transformation domain has useful and informative statistical properties, which make signal analysis simpler. In the Karhunen-Loeve transformation domain, data points (features) can be systematically selected that contain information required for accurate rhythm classification. This method of extracting data, feature selection, results in fewer data points for processing. Results of the Icarhunen-Loeve transformation with feature selection may confirm CWA to be a feasible algorithm for consideration in ATD algorithm design.