A Fractionation-based Local Activation Wave Detector for Atrial Electrograms of Atrial Fibrillation

Abstract

Atrial fibrillation (AF) electrograms (EGMs) present high heterogeneity of morphologies and amplitudes that turns the detection of their local activation waves (LAWs) a very hard task to perform. In this study, a novel fractionation-based LAW detector for bipolar EGMs is introduced. The method modifies traditional Botteron's band-pass filtering, decreasing its low cutoff frequency from 40 to 20 Hz, thus benefiting slow local activations detection. Furthermore, high and low amplitude activations in complex fractionated atrial electrograms (CFAEs) are equalized, thus facilitating the detection of low amplitude activations. A minimum refractory period of 50 ms is imposed between activations, however, additional minor activations are later sought for intervals longer than the median cycle length. All the LAWs from a set of 40 real bipolar EGMs, mostly CFAEs, were manually annotated by expert physicians and served to evaluate performance. Detections closer than 40 ms to a manual annotation were considered as correct. Detection results provided 95.41%, 92.13% and 96.37% in Sensitivity, Accuracy and Precision for CFAEs, respectively, whereas for less fractionated EGMs they were 100% in any case. Therefore, the new LAW detector has provided robust performance even under highly fractionated EGMs.

Comparative Study of Methods for Atrial Fibrillation Cycle Length Estimation in Fractionated Electrograms

“…The HFD method based the detection process on a hybrid approach of a core amplitude detection, using both a blanking period between activations of 50 ms and a fixed amplitude threshold, as well as on a later detection of lower activations within intervals longer than the median CL [10]. The preprocessing performed is also similar to Botteron's, using the same implementation of the 20 Hz low-pass filter, so just the band-pass filter low cut-off is decreased from 40 down to 20 Hz, since it has been observed that, with this configuration, those activations with a predominance of slow components were better processed, easing the later detection stage.…”

“…The detection stage begins with the main amplitude detection, which is facilitated by performing a reduction of the characteristic wide amplitude variability in CFAEs, which are selected as a function of the fractionation degree of each EGM [10]. After the amplitude detection, the median CL of the EGM is calculated from the current set of detected activations and taken as a reference value to analyze those intervals between activations longer than it.…”

“…A reduction of the amplitude threshold proportional to the length of each interval longer than the median CL is applied with the aim to detect low amplitude activations that could have been lost. Finally, aiming to improve the detection precision and a further better estimation of the CL, the activation times are set on the barycenter of each LAW, computed as the value that divides the area of the modulus of the LAW into two equal parts [10]. Furthermore, should the estimation of the barycenter of two adjacent activations is located within an interval shorter than 50 ms, the activation with the lowest amplitude is discarded from the set of detected activations [10].…”

“…For this comparative study, two reference methods were employed: the adaptive amplitude threshold (AAT) method developed by Faes et al [8] and the cycle length iterator (CLI) method introduced by Ng et al [9]. In addition, a third novel LAWs detector has been introduced making use of a hybrid fractionation degree (HFD) approach [10]. The method performs LAW detection by considering both amplitude and CL of local waves together with an estimation of the fractionation degree of the EGM under study.…”

Discrimination Between CFAEs of Paroxysmal and Persistent Atrial Fibrillation with Simple Classification Models of Reduced Features

“…The AFCL was measured by detecting local activation waves within the CFAEs [16], inverting the distances between consecutive activations and averaging them. The AFCL was computed on the 16 s-length CFAEs at all the recording sites.…”