Deconvolution and causality for 1:1 arrhythmia discrimination in dual chamber defibrillator

Automatic recognition of 1:1 ventricular tachycardia (VT) from supraventricular tachycardia (SVT) in dual chamber implantable cardioverter defibrillator (ICD) is still an open issue. A new criterion is proposed, based on the modelling of a causal versus an anti-causal relationship between the atrial and the ventricular ICD stored electrograms (EGM). A cardiac activation model is presented, consisting on (a) a pulsed activation in the atria, (b) two impulse responses modelling the atrial and the ventricular depolarization, and (c) a filter describing the atrio-ventricular (AV) propagation delay. A minimum absolute error algorithm is developed, basing on template generation, deconvolution, stochastic gradient descent and projection onto convex sets. Examples of the algorithm when fitting the model to sinus rhythm (SR), SVT and VT records are employed. The algorithm is shown to be robust in the delay-filter order choice. We propose this method as a promising framework to design low computational burden 1:1 SVT-VT discrimination algorithms.