Extracting atrial activations from intracardiac signals during atrial fibrillation using adaptive mathematical morphology

Abstract

The detection of intracardiac activities is a major issue in the processing of atrial fibrillation signals. we evaluate a method based on mathematical morphology with an adaptive structuring element in order to extract the atrial activations from intracardiac electrograms. The structuring element is continuously updated for each activation based on the morphological characteristics of the previously detected activations. A dataset of recordings from patients with chronic atrial fibrillation who underwent catheter ablation were used in order to evaluate the performance of the proposed method. Results show high performance compared to a dataset manually annotated by an expert. The detection rate, sensitivity and positive prediction value (PPV) were respectively 99.1% ,99.5%, 99.5%. The proposed method is fast and offers low computational cost, which makes it a suitable approach for realtime/online scenarios. IntroductionIntracardiac activation-time detection algorithms can be used in atrial fibrillation (AF) electrogram analysis as a first step in estimating AF characteristics. Characteristics such as cycle lengths (AFCLs) of atrial activations (AAs) can help predict persistent AF ablation outcomes [1,2]. Moreover, AFCLs can be used to determine sites with high frequency activities which may help identify critical ablation targets to restore sinus rhythm [3].However, the varying amplitude and morphologies of AA during AF makes AA extraction difficult. Through a time consuming task, one can always measure activation intervals manually by using calipers and then averaging several measurements to find the mean AFCL. Alternatively, automatic detection methods can be used to extract AAs even though most are limited due to the use of various thresholds [4]. Spectral analysis can also help to determine AFCL by extracting AF dominant frequency (DF), i.e the frequency that has the maximum power [5]. DF methods try to approximate AA rates by a single sinusoid. Therefore, by nature, these methods are not suitable when AF electrograms present irregularities. A time domain iterative method has been proposed [6] to extract AAs from intracardiac electrograms (ICEGs). The detection threshold level is iteratively adjusted until the mean and the median of AFCL converge on a signal segment.The purpose of this study is to evaluate a new adaptive mathematical morphology (MM) approach that extracts AAs based on their morphological features. The structuring element used in this method is adapted after each AA extraction in order to have a more reliable extraction. Methods Patients and Data AcquisitionStudy Population. The study group for this research consists of three consecutive patients (63 ± 1 years) with chronic AF (sustained AF duration 17 ± 8 months) who underwent catheter ablation.Electrophysiological Study. The following catheters were introduced via the femoral veins: 1) a 3.5 mm cooled-tip catheter for mapping and ablation (Navistar ® Thermocool ® , Biosense Webster ® ); and 2) a circumferential duodecapolar Lasso ® c...

Evaluation and optimization of novel extraction algorithms for the automatic detection of atrial activations recorded within the pulmonary veins during atrial fibrillation

“…The adaptive mathematical morphology (AMM) algorithm has been thoroughly presented in previous papers aiming at either extraction of QRS complexes from the surface ECG [ 17 ] or detection of AA sequences from intracardiac signals [ 21 ]. Briefly, the AAs are extracted using a structuring element (SE) which is continuously updated for each new AA based on the topological features of the previous detected AA.…”

Experimental Study of Atrial Fibrillation Cycle Length During Rapid Atrial Septal Pacing

“…One dipole neighbourhood was considered captured if the local CL was within ±5% of the PCL. AFCL was measured on bipolar electrograms using a nonlinear filtering technique that uses short-and long-term electrogram energies for robust AFCL extraction [10,11]. Sliding short-and longterm signal energies were measured for each sample in the EGM.…”