Screening for paroxysmal atrial fibrillation using atrial premature contractions and spectral measures

Abstract

We present a technique for screening for imminent onset of paroxysmal atrial fibrillation (PAF) through automated assessment of 30-minute segments of electrocardiogram (ECG), which do not contain any episodes of atrial fibrillation. Algorithmic development was carried out using a training database of 75 half-hour records drawn from two subject groups. Subjects in the first group provided segments with PAF in the five minutes after the 30-minute recording; the second group do not have PAF (control subjects or subjects with non-PAF cardiac pathology). To differentiate between pre-PAF segments and non-PAF segments a linear discriminant classifier was developed, using the number of Atrial Premature Contractions (APCs) and two spectral measures as features. An independent test set of 72 recordings (28 pre-PAF and 44 non-PAF) was then classified, with an accuracy of 75% (sensitivity 79%, specificity 72%). When tested against a second database of subjects with no known cardiac pathology, the specificity rose to 95%.