Knowledge-based system for classification of body surface potential maps

A rule-based expert system shell was developed in order to correctly classify 159 normal subjects (N), 103 patients with anterior myocardial infarction (AMI), 130 patients with inferior myocardial infarction (IMI) and 116 patients with left ventricular hypertrophy (LVH). Input variables were instantaneous voltage measurements obtained from 120 simultaneously recorded electrocardiographic leads by sampling the time-normalized P, PR, QRS and STT waveforms at equal intervals; this resulted in 8, 8, 18 and 18 samples for P wave, PR segment, QRS and STT waveforms, respectively. Added to these variables were the actual durations of these waveforms. Linear discriminant functions (LDF) were computed for each of the 6 bigroup comparisons (N vs. AMI, N vs. IMI, AMI vs. IMI, etc.). The strategy for group-assignment was implemented in a tree structure of metarules containing 42 nodes and using a set of rules to guide its problem-solving activity. Tests revealed correct classification rates of 92%, 94%, 93% and 89% for N, AMI, IMI and LVH, respectively