Nonlinear statistical methods for atrial fibrillation detection on electrocardiogram

The rapid and disorganized electrical activity that characterizes Atrial Fibrillation generates an increases in both variability and complexity of RR intervals series. Based on these characteristics which are variability and complexity, we have developed an algorithm combining two nonlinear statistical techniques in order to detect the presence of Atrial Fibrillation. These two nonlinear statistical methods are: Poincare plot which quantifies the variability of RR intervals series and the sample entropy which characterizes the complexity of the RR intervals series. We used the MIT-BIH Atrial Fibrillation database to train the algorithm and then we tested it on the MIT-BIH Arrhythmia database. Using thresholds and segment length determined by Receiver Operating Characteristic (ROC) curves we achieved Se=97.01% and Sp=93.92% for the MIT-BIH Atrial Fibrillation database and we obtained Se=98.67% and Sp=86.14% for the MIT-BIH Arrhythmia database. The proposed algorithm is compared to another method, namely “Dash’s” detection method [1]. For the Dash’s method, we achieved Se=97.09% and Sp=90.98% for the MIT-BIH Atrial Fibrillation database and we obtained Se=96% and Sp=82.31% for the MIT-BIH Arrhythmia database.