Classification of pathologies by reduced sequential potential maps

Abstract

Body surface potential mapping (BSPM) is an electrocardiographic measuring technique which produces the data as a series of three-dimensional maps. These maps are assumed to contain information which may help classify subjects for diagnostic purposes more effectively than standard ECGs. As quantitative classification of the complete sequences of maps is complex and cumbersome, the present study uses extracted features which characterise the data. The features, which have been presented and evaluated in a recent work, have been extracted after the maps were processed by a compression technique which conserved the spatial details of the maps. The compression by two-level thresholding converted the sequences of maps into sequences of annuli, from which the following features were extracted: time indices, velocity vector magnitude, loci in three-dimensional space of the centres of mass and cross-correlation coefficients between successive annuli in the sequence. Here, three different classification methods are applied to these features: statistical methods, the Fisher linear discriminant method and visual inspection. BSPMs from 54 subjects are used: 25 normal, 11 WPW syndrome and 18 CAD cases. It is found that by applying a decision role which comprises all features, the procedure offers a completely accurate classification of the subjects to their groups. The three-dimensional centre of mass is found to be the single best classifier; successfully categorising 20/25 of the normals 17/18 of the CAD patients and 11/11 of the WPW patients.