Nonparametric identification of discriminative information in body surface maps.

A nonparametric method, based on the Kolmogorov-Smirnov (K-S) test was used to detect significant differences between classes of body surface maps (BSPM's). By systematic application of the method throughout the cardiac cycle, discriminative spatio-temporal information can be identified. In a second method, a Sebestyen linear transformation (SLT) was derived to give estimates of pairwise, linear separability of clinical classes. The utility of the method was illustrated by the pairwise comparison of 40 normal subjects (NOR), 40 patients with anterior myocardial infarction (AMI), and 40 with inferior myocardial infarction (IMI). The application examples demonstrated that: a) diagnostic information in low potential amplitude regions may surpass that in high amplitude regions, b) probability distributions of characteristic features showed small overlap in NOR versus AMI and NOR versus IMI dichotomies although they were not linearly separable, and c) the single best separating potential sample in the K-S sense for NOR versus AMI or NOR versus IMI dichotomies recovered 88 and 73% of the SLT performance, respectively.