A new algorithm for short term prediction of persistent atrial fibrillation

Abstract

Atrial fibrillation (AF) is the most common cardiac arrhythmias which affects more than 2 million US adults. Paroxysmal AF is characterized by recurrent AF episodes that stop on their own in less than 7 days. If the AF episodes last for more than 7 days, it is unlikely that they will stop on their own, and they are then known as persistent AF episodes which necessitates treatment with pharmacological or electrical cardioversion. This paper develops a new algorithm for short term prediction of persistent AF episodes. The proposed data-driven model is optimized with respect to predictions of persistent atrial fibrillation using weighted support vector machines and cost-sensitive learning. The proposed prediction model can be further personalized to assist clinicians to deliver proactive treatment therapies that can prevent persistent AF episodes from occurrence.