Evaluation of lead selection methods for optimal reconstruction of body surface potentials

In this study, several methods for optimal lead selection from multilead electrocardiographic recordings are analyzed. Two different lead selection methods have been implemented. For their evaluation, a linear transformation that reconstructs nonselected leads from selected leads is computed according to the least squares optimization, and the performance is evaluated in terms of the mean square error of the derived potentials and correlation. The algorithms were tested on a database of 72 body surface potential recordings: 18 controls, 18 bundle-branch block, 18 myocardial infarction, and 18 ventricular hypertrophy. Each data set was divided into a study and test subsets. Two experiments were carried out: (1) The lead selection, transformation matrix, and performance evaluation is carried out over the test data set (ideal case), and (2) the lead selection and transformation matrix is carried out over the study data set, but the performance is evaluated over the test data set (real case). Our results show important reconstruction errors with either lead selection methods, and only increasing the number of leads reduces the error in reconstruction. However, if a reduced number of leads are to be selected outside the standard 12-lead electrocardiogram, the method proposed by Lux has been shown to be the best option.