Improving Atrial Fibrillation Detection Using Adaptive Template Matching

We develop and assess algorithms for atrial fibrillation detection in electrocardiograms acquired over long time periods. The algorithms exploit the use of adaptive templates for QRS complexes that can accommodate changes in rhythms due to factors such as measurement device movement or the onset of ventricular beats. Adaptive templates are also used to represent P wave morphology. The use of adaptive templates makes these techniques particularly valuable for automated processing in devices such as insertable cardiac monitors that support evaluation over substantial periods of time. The algorithms are compared with a baseline AF detection algorithm that is based on the consistency of RR intervals. A probabilistic model for the likelihood of AF is generated by applying logistic regression to various electrocardiogram descriptors. We use data from four MIT-BIH databases to quantify the performance of the algorithms and to show that performance is improved significantly by the use of the adaptive QRS and P wave templates.