Improved estimation of pericardial potentials from body-surface maps using individualized torso models

Clinical applicability of inferred pericardial potentials is limited because accuracy is significantly affected by noise in surface electrocardiograms (ECGs), errors in electrode location on the torso model, and errors in the geometry and inhomogeneities of the torso model itself. To quantify effects of electrode location and geometric errors in torso-surface models, we measured locations of 190 electrodes used in body-surface mapping of 11 adults, along with over 2,000 sites on each torso surface. Measurements were made to within 2 mm with an Immersion Personal Digitizer. To quantify effects of errors in pericardial-surface models we also estimated heart position, size, and orientation in each subject from ultrasonic images registered to the body-surface coordinates. Known pericardial potentials were taken from epicardial measurements made during QRS with a 90-electrode sock in an adult male undergoing cardiac surgery. Body-surface ECGs were calculated for each individual from the pericardial maps, using standard boundary-element methods. Accuracy of zero-order-Tikhonov inverse solutions was tested in 91-node pericardial and 1,026-node torso models, individualized for each subject. With 10 microv rms noise added to surface potentials, the optimal regularization constant at each instant in QRS gave a relative error of 0.44 +/- 0.03; it was 0.47 +/- 0.03 using the composite residual and smoothing operator (CRESO) technique. When calculated body-surface potentials from the first 10 subjects were placed at corresponding electrode positions on the torso of the eleventh subject, whose heart size and orientation was the mean of the other 10 subjects, relative error increased to 0.87 +/- 0.06 for optimal regularization. CRESO failed in the fixed torso model. Results demonstrate that a fixed model does not provide useful estimates of pericardial potentials, and that individualized models enhance the performance of techniques for the estimation of regularization parameters.