Support vector classification of acute myocardial ischemia with reduced sets of body surface potential map electrodes

Body surface potential maps (BSPM) can be used to localise supply ischemia [1]. The complete lead set could be used for more exact classification than with any reduced set, but for clinical purposes reduced sets are useful. In this work, three best unipolar ECG leads are determined for each pairwise classification between artery groups and non-ischemic control subjects. Most suitable ECG parameters are first selected using interpatient variance. Separating hyperplanes are then found using linear kernel support vector machine (SVM). Success in classification with the hyperplane is used as an indicator of goodness of an electrode set. The best classifying channels are then located with simulated annealing. The results indicate that in addition to close proximity to the heart, an electrode on the posterior thorax is needed for differentiating between arteries. Most of the resulting leads are found in the approximate directions of heart axes