Identification of patients at risk for ventricular tachycardia by means of body surface potential maps

In this work we attempted the stratification of patients at risk for VT by means of BSPM recorded during sinus rhythm, based on the evidence that specific arrhythmogenic alterations manifest on body surface potentials. Due to the high dimensionality of BSPM data and the limited number of available patients, a feature extraction step was necessary prior to the classifier design. Feature extraction was performed by means of linear expansions of two different time intervals: QRS and ST-T complexes. Two approaches were studied: the Karhunen-Love transform (KLT) and spatio-temporal expansions. A multivariate linear discriminant analysis was applied to the extracted features to classify the study population in two groups: VT and non-VT. Our results showed that spatio-temporal features (SE = 83%, SP = 86%) obtained similar classification results than KLT features (SE = 78%, SP = 93%) with a lower computational cost. For comparison, a method reported in the literature based on QRST integral maps was implemented, obtaining results within the same range (SE = 88%, SP = 72%).