Classification of normal and ischemia from BSPM by neural network approach

Body surface potential mapping (BSPM) techniques have improved diagnosis of ischemia patients with normal resting ECG. The authors report BSPM data from 32 normal and 24 ischemia patients (all adults). The studies included: (1) separability of normal and ischemia from QRS integral maps (represented by normalized K-L expansion) by a two-layer forward network method; (2) evaluation of classification performance versus the number of representing eigenvectors; and (3) distribution and relationship of normal and ischemia QRS integral maps. Results show that ischemia and normal were distributed very close to each other, but can be well separated. Only ten expansion coefficients were needed to reach a 98% classification rate evaluated by a resubstitution method and an 82% classification rate by a cross-validation method.<
>