Bioautomation (Dec 2005)
Fast Algorithm for Vectorcardiogram and Interbeat Intervals Analysis: Application for Premature Ventricular Contractions Classification
Abstract
In this study we investigated the adequacy of two non-orthogonal ECG leads from Holter recordings to provide reliable vectorcardiogram (VCG) parameters. The VCG loop was constructed using the QRS samples in a fixed-size window around the fiducial point. We developed an algorithm for fast approximation of the VCG loop, estimation of its area and calculation of relative VCG characteristics, which are expected to be minimally dependent on the patient individuality and the ECG recording conditions. Moreover, in order to obtain independent from the heart rate temporal QRS characteristics, we introduced a parameter for estimation of the differences of the interbeat RR intervals. The statistical assessment of the proposed VCG and RR interval parameters showed distinguishing distributions for N and PVC beats. The reliability for PVC detection of the extracted parameter set was estimated independently with two classification methods - a stepwise discriminant analysis and a decision-tree-like classification algorithm, using the publicly available MIT-BIH arrhythmia database. The accuracy achieved with the stepwise discriminant analysis presented sensitivity of 91% and specificity of 95.6%, while the decision-tree-like technique assured sensitivity of 93.3% and specificity of 94.6%. We suggested possibilities for accuracy improvement with adequate electrodes placement of the Holter leads, supplementary analysis of the type of the predominant beats in the reference VCG matrix and smaller step for VCG loop approximation.
