Diagnosis of Atrial Fibrillation Using Electrograms from Chronic Leads: Evaluation of Computer Algorithms

Publisher Website
Google Scholar
Abstract
This study compares the performance of three detection algorithms for the recognition of atrial fibrillation in chronic pacing leads. Multiple serial recordings were obtained of wideband and filtered electrograms from chronic atrial and ventricular leads in dogs for a period up to 55 days following implantation. Each dog was recorded in sinus rhythm and induced atrial fibrillation. Four days were chosen for processing: The day of implantation and a day in the first, second or third, and fifth weeks. Three signal processing methods were assessed for performance in detection of atrial fibrillation: software recognition of rate with automatic threshold control, amplitude distribution, and frequency spectral analysis. A software trigger for rate determination was adjusted to thresholds of 10, 20, and 30% of maximum baseline-to-peak amplitude. At 10%, a rate boundary anywhere between 420 and 560 beats per minute (bpm) perfectly separated atrial fibrillation from sinus rhythm even though atrial electrograms were contaminated with large QRS deflections and double-sensing was present. At 20% and 30%, a rate boundary around 300 bpm could be used, but sensitivity and specificity were reduced to 90%. In amplitude distribution analysis, a percent of time within a baseline window provided perfect separation of atrial fibrillation from sinus rhythm. In all cases, the signal was within this window Jess than 43% of the time in atrial fibrillation, and more than 43% in sinus rhythm. In spectral analysis, frequency bands were examined for power content. In the 6 to 30 Hz band atrial fibrillation contained the greater power. Choosing 58% of total power as a discriminant, sensitivity and specificity of atrial fibrillation detection were 100% and 95% respectively.

This publication has 14 references indexed in Scilit: