Computer Diagnosis of Biventricular Hypertrophy from the Orthogonal Electrocardiogram

Abstract

Orthogonal electrocardiograms obtained from 87 patients with necropsy evidence of biventricular hypertrophy (BVH) were compared to record samples obtained from normal subjects (N) and patients with left or right ventricular hypertrophy (LVH or RVH). From 333 different ECG measurements an attempt was made to select optimal discriminators of the various pathological entities. The best separation between the BVH sample and other entities was obtained by utilizing linear discriminant function analysis and a likelihood ratio test. By using a combined covariance matrix of BVH versus N, LVH, and RVH, 69% of the BVH sample was correctly classified, thus demonstrating that multivariate analysis can lead to a substantial improvement in diagnostic classification over previous studies reported in the literature. To test the multivariate classification method against completely independent record samples, new series of LVH and RVH cases were compared to the BVH sample. The results were similar to those obtained against the original samples. When a "clinical" BVH sample was classified by the combined covariance matrix, 44% of the new cases were classified correctly. A search was also made for optimal scalar and vectorial measurements that can be used in routine ECG interpretation without access to computer facilities. On using 96 percentile ranges, the separations were less efficient than those obtained by multivariate analysis but they still compared very favorably with previous reports.