Spatiotemporal model-based estimation of high-density atrial fibrillation activation maps

Abstract

Examination of activation maps using multi-electrode array (MEA) sensors can help to understand the mechanisms underlying atrial fibrillation (AF). Classically, creation of activation maps starts with detection of local activation times (LAT) based on recorded unipolar electrograms. LAT detection has a limited robustness and accuracy, and generally requires manual edition. In general, LAT detection ignores spatiotemporal information of activation embedded in the relation between electrode signals on the MEA mapping sensor. In this work, a unified approach to construct activation maps by simultaneous analysis of activation patterns from overlapping clusters of MEA electrodes is proposed. An activation model fits on the measured data by iterative optimization of the model parameters based on a cost function. The accuracy of the estimated activation maps was evaluated by comparison with audited maps created by expert electrophysiologists during sinus rhythm (SR) and AF. During SR recordings, 25 activation maps (3100 LATs) were automatically determined resulting in an average LAT estimation error of -0.66±2.00 ms and a correlation of ρs=0.98 compared to the expert reference. During AF recordings (235 maps, 28226 LATs), the estimation error was -0.83±6.02 ms with only a slightly lower correlation (ρs=0.93). In conclusion, complex spatial activation patterns can be decomposed into local activation patterns derived from fitting an activation model, allowing the creation of smooth and comprehensive high-density activation maps.