Improved EASI coefficients: their derivation, values, and performance

The EASI lead system, which is based on the dipole hypothesis of vectorcardiography, offers the possibility of deriving the standard 12-lead electrocardiogram (ECG) and other desired leads from ECGs recorded at only 4 sites; it uses the Frank E, A, and I electrode locations, a fourth electrode location (S) at the manubrium, and a reference electrode. Accordingly, the electrodes of this system can be applied rapidly on easy-to-locate, stable anatomical sites that leave the precordium free for other diagnostic procedures. In early EASI implementations, the derived leads differed from actual leads by more than some clinicians found acceptable. As these differences were thought to be caused by the fact that the coefficients that were used had been derived from a limited data set, we have calculated a new set of EASI coefficients for the standard 12 leads, and several other leads, by using a data set of 983 adult subjects with 120-lead ECGs and well-documented diagnoses. This database is a concatenation of 2 previously described ones: one consisting of 892 persons (normal subjects, postmyocardial-infarction patients with and without arrhythmias, and patients with ventricular arrhythmias but no history of myocardial infarction) and the other consisting of 91 patients with single-vessel coronary artery disease who underwent coronary balloon-inflation angioplasty. In addition to the coefficients for the standard 12 leads (derived for standard limb leads as well as for Mason-Likar leads), we derived coefficients for six additional unipolar leads (posterior V(7)-V(9), and right-sided V(3)R-V(5)R), the Frank orthogonal leads, and three bipolar, vessel-specific leads that have been previously shown to exhibit optimal sensitivity for acute myocardial ischemia. We also derived coefficients for the modified electrode locations of the EASI system that must be used with patients who have undergone a midline sternotomy. Optimal coefficients for lead transformations were determined by maximizing the ensemble average (over the entire data set) of the correlation between the derived and the true lead for the chosen interval of the averaged complex. For derived standard limb leads, the amplitude was adjusted to give the best root-mean-square fit over the entire PQRST interval, whereas for derived Mason-Likar leads it was adjusted to give the best ST-segment fit. The entire set of coefficients and their corresponding goodness-of-fit measures are presented.