Single-Feature Method for Fast Atrial Fibrillation Detection in ECG Signals

Abstract

Atrial fibrillation (AF) is the most common arrhythmia in adults and is associated with a higher risk of heart failure or death. Here, we introduce simple and efficient method for automatic AF detection based on symbolic dynamics and Shannon entropy. This method comprises of three parts. Firstly, QRS complex detection is provided, than the raw RR sequence is transformed into a sequence of specific symbols and subsequently into a word sequence and finally, Shannon entropy of the word sequence is calculated. According to the value of Shannon entropy, it is decided, whether AF is present in the current cardiac beat. We achieved sensitivity Se=96.32% and specificity Sp=98.61% on MIT-BIH Atrial Fibrillation database, Se=91.30% and Sp=90.8% on MIT-BIH Arrhythmia database, Se=95.6% and Sp=80.27% for Long Term Atrial Fibrillation database and Se=93.04% and Sp=87.30% for CinC Challenge database 2020. The achieved results of our one-feature method are comparable with other authors of more complicated and computationally expensive methods. Our ECG experts found that public databases contain errors in annotations (in sense of AF). It means that results are affected by errors in annotations. Many errors were found in Long-Term AF database, several also in MIT-BIH AF database and MIT-BIH Arrhythmia database. Testing algorithms on poorly annotated databases cannot bring reliable results and algorithms useful in real medical practice. The examples of such annotations are reported in this study.