Optimal lead configuration in the detection and subtraction of QRS and T wave templates in atrial fibrillation

An in-house written software package was used to generate templates of QRS complexes and T waves using self-similarity techniques, and leave a residual signal representing only atrial activity. The templating performance differences between individual ECG leads were tested. The mean number of templates in each of the 3 ECG leads was 8/spl plusmn/4, 2/spl plusmn/1, 7/spl plusmn/5 for the QRS complexes, and 11/spl plusmn/5, 11/spl plusmn/8, 10/spl plusmn/5 for the T waves. Lead 2 generated less QRS templates than 2 other leads, left less unmatched QRS complexes and T waves (all p<0.005). The average correlation coefficients between templates were 0.69/spl plusmn/0.27, 0.21/spl plusmn/0.36, and 0.63/spl plusmn/0.27 respectively for QRS complexes, and 0.33/spl plusmn/0.24, 0.46/spl plusmn/0.31, and 0.50/spl plusmn/0.24 respectively for T wave in leads 1, 2, and 3. These data suggest that the QRS subtraction process performed well. Lead 2 (a bipolar V5) performed best, but optimal subtraction may require more templates, and hence auto-tuning by the algorithm to the signal encountered.