Identification of Strict Left Bundle Branch Block, Using a Moving Dipole Model

Werner Bystricky
TSD, Langenburg


Abstract

Aim: This study did investigate how well a moving dipole approximation of the electrical activity of the heart can contribute in differentiating strict left bundle branch blocks (sLBBB) from other unspecified depolarization abnormalities. Methods: Training data were 300 standard 12 lead ECGs with given sLBBB diagnoses from the LBBB initiative of the ISCE 2018 meeting. A moving dipole model was applied to the heart beats lying within the central 5 seconds of each ECG, providing the 3-dimensional time courses of the dipole momentum and of the dipole location. Furthermore, vectorcardiographic representations of the same heart beats were calculated using the inverse Dower transformation, the Guldenring transformation, and the Kors transformation. The individual transformed 3-dimensional signal data were averaged over time segments of 10, 20, and 40 ms length, starting at the QRS onset and ending 160 ms later. These averaged signal data of a given transformation were used as input for a logistic regression model. The model size was reduced by eliminating non-significant input parameters through a genetic parameter selection algorithm. The classification performances of the various input dataset were compared by calculating the area under the receiver operating characteristic curve (AUC) in a cross validation approach. Results: Averaging the transformed signals over 10 ms time segments performed better than averaging over 20 ms or 40 ms time segments. This was true for each investigated transformation setting. The classification performances of the three linear vectorcardiographic transformations (inverse Dower, Guldenring, and Kors) were very similar (AUC = 0.81). The best classification performance was reached by using both, the momentum and the location of the moving dipole as input for the logistic regression model (AUC = 0.85). Conclusion: The location information about the heart’s electrical activity, revealed by the moving dipole model, enhances identification of sLBBB.