Diagnostic classification of patients with ventricular tachycardia, based on spatial and temporal features derived from body-surface potential maps

The authors evaluated the ability of spatial features derived from electrocardiographic QRST-area distributions, and of temporal electrocardiographic measures (heart rate, QRS duration and corrected QT interval), to identify patients at risk for ventricular arrhythmias. Electrocardiograms from 120 leads were recorded simultaneously during sinus rhythm for 102 patients who had had ventricular tachycardia (VT) and for 102 patients who had had a myocardial infarction (MI) but no history of arrhythmias. The Karhunen-Loeve (K-L) transform was used to reduce the QRST-integral maps to 16 coefficients. The best features for discriminating between the two groups were selected by stepwise discriminant analysis, and bootstrap method was used to estimate diagnostic performance on a prospective population. With a set of 8 K-L features, the test-set sensitivity was 90.3/spl plusmn/4.3% and specificity was 78.0/spl plusmn/6.1%. When QRS duration was added as a supplementary feature to the 8 K-L coefficients, specificity increased to 80.9/spl plusmn/5.4%. All three temporal features are highly correlated among themselves; therefore, only one suffices to supplement K-L coefficients.