The Impact of Filter Settings on Morphology of Unipolar Fibrillation Potentials

Using unipolar atrial electrogram morphology as guidance for ablative therapy is regaining interest. Although standardly used in clinical practice during ablative therapy, the impact of filter settings on morphology of unipolar AF potentials is unknown. Thirty different filters were applied to 2,557,045 high-resolution epicardial AF potentials recorded from ten patients. Deflections with slope ≤ − 0.05 mV/ms and amplitude ≥ 0.3 mV were marked. High-pass filtering decreased the number of detected potentials, deflection amplitude, and percentage of fractionated potentials (≥ 2 deflections) as well as fractionation delay time (FDT) and increased percentage of single potentials. Low-pass filtering decreased the number of potentials, percentage of fractionated potentials, whereas deflection amplitude, percentage of single potentials, and FDT increased. Notch filtering (50 Hz) decreased the number of potentials and deflection amplitude, whereas the percentage of complex fractionated potentials (≥ 3 deflections) increased. Filtering significantly impacted morphology of unipolar fibrillation potentials, becoming a potential source of error in identification of ablative targets. Graphical Abstract Impact of filtering on morphology of unipolar AF potentials. High-pass, low-pass and notch filters were applied to 2,557,045 high-resolution epicardial AF potentials recorded from ten patients. Filtering significantly impacted AF potential morphology, i.e., number of detected potentials, peak-to-peak amplitude, number of deflections, and fractionation delay time. CFP, complex fractionated potential (≥ 3 deflections); DP, double potential (two deflections); FDT, fractionation delay time; SP, single potential (one deflection). Electronic supplementary material The online version of this article (10.1007/s12265-020-10011-w) contains supplementary material, which is available to authorized users.


Introduction
Although insight into the pathophysiologic basis of fractionated atrial electrograms has increased in the past years, its "fact or artifact" remains a topic of debate. Extracellular electrograms-recorded directly from the heart-are generated by depolarization of cardiomyocytes, so signal morphology could provide information about the electrophysiological characteristics of the underlying myocardium. [1] Atrial potentials consisting of multiple components (i.e., fractionated) h a v e b e e n l i n k e d t o a b n o r m a l c o n d u c t i o n a n d arrhythmogenicity in patients with atrial fibrillation (AF) [1,2], which led to targeted ablation of such fractioned potentials. However, the link between morphology of atrial potentials and pathology has proven to be anything but straightforward, Associate Editor Sunil Kapur oversaw the review of this article Electronic supplementary material The online version of this article (https://doi.org/10.1007/s12265-020-10011-w) contains supplementary material, which is available to authorized users.
considering that multiple physiological mechanisms and measurement properties, such as filtering, can cause fractionated potentials as well [3,4]. In the 1980s, Waxman and Sung discovered the phenomenon of frequency-dependent fractionation in human bipolar ventricular electrograms [5]. Klitzner and Stevenson showed that increasing the high-pass filter frequency above 10 Hz decreased potential duration and amplitude, whereas low-pass filtering altered potential amplitude slightly if decreased up to 100 Hz [6]. Such low-and highpass filters are commonly used in clinical practice, as well as utilization of a 50 Hz (or 60 Hz) notch filter to suppress power-line interference.
Despite the clinical failure of targeting complex fractionated atrial electrograms (CFAEs) as stand-alone therapy [7][8][9], using atrial electrogram morphology as guidance for ablative therapy is regaining interest [10]. Up till recently, most clinically used ablative systems preferred bipolar above unipolar measurements, given its ability to reduce far-field potentials [11]. However, bipolar signals fail to represent incoherent waves during AF, which is why currently emerging innovative ablative systems, such as RhythmView™ and AcQMap®, prefer unipolar electrograms to identify local activation [12,13]. As the study of van der Does et al. demonstrated electrogram morphology at the epi-and endocardium to be comparable [14], epicardial electrograms are suitable to investigate signal morphology, particularly as direct contact between the electrode and atrial tissue can be assured.
To our knowledge, the impact of filtering on unipolar fibrillation potentials has never been investigated in humans. This study therefore aims to elucidate the consequences of high-pass, low-pass, and notch filtering on unipolar fibrillation potentials in AF patients.

Study Population
The study population consisted of ten adult patients with a history of paroxysmal or persistent AF undergoing elective open-heart mitral valve surgery in the Erasmus Medical Center Rotterdam. This study was approved by the institutional medical ethical committee (MEC 2010-054/MEC 2014-393) [15,16]. Written informed consent was obtained from all patients. Patient characteristics (e.g., age, medical history, date of AF diagnosis) were obtained from the patient's file.

Mapping Procedure
Epicardial high-resolution mapping was performed prior to commencement to extra-corporal circulation, as previously described in detail [17][18][19]. A temporal bipolar epicardial pacemaker wire attached to the RA free wall served as a reference electrode. A steel wire fixed to the subcutaneous tissue of the thoracic cavity was used as an indifferent electrode. Epicardial mapping was performed with a 192electrode array (electrode diameter 0.45 mm, interelectrode distances 2.0 mm). The right pulmonary vein (PV) area was mapped from the sinus transversus fold along the borders of the right pulmonary vein down towards the atrioventricular groove (as illustrated in the left panel of Fig. 1). Ten seconds of AF were recorded, including a surface ECG lead, a calibration signal of 2 mV and 1000 ms, a bipolar reference electrogram, and all unipolar epicardial electrograms. Data was stored on a hard disk after amplification (gain 1000), filtering (band-pass 0.5-400 Hz), sampling (1 kHz), and analog to digital conversion (16 bits).

Filter Settings
The impact of additional high-pass, low-pass, and notch filtering (i.e., narrow band-stop filter), as illustrated in the middle panel of Fig. 1 Settings were based on frequently used filter options within clinical mapping systems. Signals were zero-phase filtered with IIR Butterworth low-and high-pass filters (2nd-order: 12 dB/octave roll-off) and/or IIR notch filter with a quality factor of 30. Bode plots of the three filters are illustrated in ESM 1.

Data Analysis
Electrogram morphology was semi-automatically analyzed in custom-made Python 3.6 software. Deflections of atrial potentials were automatically marked if the slope was ≤ − 0.05 mV/ ms and the amplitude ≥ 0.3 mV; the refractory period was set to 40 ms [20]. The steepest negative deflection of a potential was classified as the primary deflection and marked as the local activation time (LAT), whereas-in case of a fractionated potential-additional deflections were classified as secondary deflections. Electrograms with injury potentials and artifacts were excluded from analysis by manual assessment and consensus of two independent investigators.
For each different filter setting, peak-to-peak amplitude (voltage), fractionation (f, number of deflections), and fractionation delay time (FDT, time interval between first and last deflection) of atrial potentials were analyzed. Peak-to-peak amplitude was analyzed for all deflections, as well as for primary and secondary deflections separately. FDT was only derived for fractionated potentials (f ≥ 2). For each morphology parameter median values of all atrial potentials within the 192-array were derived and compared between filter settings. In addition, each fibrillation potential was classified as either single potential (SP, f = 1), double potential (DP, f = 2), or complex fractionated potential (CFP, f ≥ 3). Figure 1

Statistical Analysis
The impact of filtering on characteristics of unipolar fibrillation potentials, including amplitude, fractionation, and FDT, was analyzed using linear mixed-effect models, while accounting for clustered data within a patient. Analyses were done for the three different types of filtering separately: lowpass, high-pass, and notch filtering. The basic model only included a random intercept and presumed no relation between filtering and morphology characteristics. Based on the Akaike Information Criterion it was checked whether addition of a random slope improved the model. To model any nonlinearity two splines were used. Residual plots were reviewed and log or square root transformed data was used if deviations from normality were observed. Statistical significance was tested using the likelihood ratio test. A p value < 0.05 was considered statistically significant. All statistical analyses were performed using R Statistical Software (RStudio, Inc., Boston, MA; version 1.0.153).

Results
Patient characteristics (n = 10) are shown in Table 1. All patients had a history of AF (paroxysmal n = 4, persistent n = 6) and age ranged from 56 to 77 years. In total, 3000 s of AF The impact of frequently used high-pass, low-pass, and notch filter settings on morphology of one example of a fractionated fibrillation potential is illustrated in Fig. 2. General consequences of all filter settings are discussed in the sections below.

Impact of High-Pass Filtering on Morphology of Fibrillation Potentials
Increasing the high-pass filter frequency had a negative impact on the number of detected fibrillation potentials and deflection amplitudes, as illustrated in Fig. 3. In the entire study population, the percentage of detected  This negative impact was also observed for median primary and secondary deflection amplitudes separately (p < 0.01), and was primarily caused by loss of high amplitude deflections, as illustrated in the histograms in the lower panel of Fig. 3 (obtained from one representative patient). The impact of high-pass filtering on the degree of fractionation, i.e., FDT and distribution of the different fibrillation potential types, is illustrated in Fig. 4. As observed in the left panel of Fig. 4, increasing the high-pass filter frequency from 0.5 to 100 Hz resulted in a slightly increasing percentage of SPs (from 36.1-57.6% to 31.6-70.2%) at the cost of (primarily) CFPs (from 15.9-36.0% to 7.3-   stacked bar-plots of median secondary deflection amplitude of one patient. For both lower figures, the data of patient 1 was taken as a representative case for all patients. The dotted vertical lines represent the median value of the corresponding stacked bar-plot, representing low-pass filtering at 400, 200, 100 or 50 Hz a trend that was also observed within primary and secondary deflections separately. This increase in deflection amplitude is primarily caused by a loss of low-amplitude deflections, as observed in the lower panel (histograms of one representative patient).
In Fig. 6, the impact of low-pass filtering on fractionation is visualized. The left upper panel illustrates a rapid decline of the percentage fractionated potentials (p < 0.01). Decreasing the low-pass filter frequency increased the percentage of SPs (from 36.1-57.6% to 100%) at the cost of both DPs (from 25.8-32.9% to 0%) and CFPs (from 15.9-36.0% to 0%). For all patients, the impact of low-pass filtering on DPs was mainly when filtering at 150 Hz or lower, whereas the presence of

Impact of Notch Filtering on Morphology of Fibrillation Potentials
As indicated in Table 2, notch filtering slightly decreased deflection amplitude and increased the percentage of CFPs, whereas no effect on FDT and the percentage of SPs and DPs was observed. Applying a notch filter at 50 Hz induced a loss of~1% detected fibrillation potentials (from 9234-14,545 to 9234-14,405 potentials, p = 0.01). This significant, yet minimal, effect was also observed in a decrease in median deflection amplitude (from 0.59-0.96 mV to 0.59-0.95 mV, p < 0.01). Median FDT was not affected by notch filtering (from 16.0-25.0 ms to 16.0-25.0 ms, p = NS). Notch filtering also had no effect on the percentage of SPs and DPs (from 36.1-57.6% to 35.8-57.6% and from 25.8-32.9% to 25.8-32.3% respectively, p = NS), but did increase the percentage of CFPs significantly (from 15.9-36.0% to 16.2-37.0%, p = 0.0157).

Filtering and Detection of Local Activation Time
As a subanalysis, the impact of filtering on detection of LAT was determined. Results are described in detail in the ESM 3. All filter settings, i.e., all high-pass, low-pass, and notch filter settings, evoked changes in LAT timing (ESM 2). Especially with more aggressive low-pass filtering, the percentage of fibrillation potentials that had a shift in LAT was high (e.g., 58.32 to 63.12% at 100 Hz). Filtering impacted LATs of all potential types (i.e., SPs, DPs, and CFPs). Nevertheless, more complex and long fractionated potentials had a greater ΔLAT-and thus shifted more-than potentials with simpler morphology.

Discussion
All clinically used mapping systems-both unipolar and bipolar-standardly use signal filters while ablating. Although all mapping systems have different default filter settings, operators have the freedom and ability to change the filter settings according to their wishes. The results of our study clearly show that filtering choices have a significant impact on unipolar signal morphology. Attempts to correct for noise or baseline drift can therefore easily result in erroneous (under)detection of fractionation and/or low-voltage areas and thus ablative targets during mapping.
Our study thereby complements and verifies previously reported findings by Schneider et al. [21] and Lin et al. [22], in which unipolar endocardial peak-to-peak voltage decreased with increasing high-pass filtering in patients with ectopic atrial tachyarrhythmias and atrial flutter, respectively. Table 2 Impact of notch filtering on unipolar fibrillation potential morphology. Delta values (Δ = notch filter on-notch filter off) for all parameters and corresponding p values based on the likelihood ratio test are given (* < 0.05). Potential type is either single (one deflection), double (two deflections), or complex (three or more deflections).

Current Clinical Use of Filtering in Ablative Techniques
Although initially developed as a stand-alone strategy, ablation of tissue exhibiting CFAEs is nowadays generally used as adjuvant therapy to PV isolation, particularly in persistent AF patients [2,23]. Pathophysiological mechanisms of CFAE include pivoting points, inhomogeneous conduction, functional conduction block, reduced cell coupling, and interstitial fibrosis [2,24]. On signal level, fractionation is often considered high-frequency content with a low-amplitude [25,26]. Accordingly, lowering particularly the low-pass filter frequency impacted the presence of (complex) fractionated potentials in our study, since high-frequency content was eliminated. So while the decision to change the low-pass filter setting during CFAE ablation is probably made to reject high-frequency noise, detection of CFPs and thereby ablative targets is also strongly impeded. As an example, low-pass filtering at 100 Hz already eliminated up to 40% of fibrillation potentials and reduced the presence of CFPs from 15.9-36.0% to 2.3-10.7% in all patients. Interestingly, increasing high-pass filtering did significantly decrease the percentage of fractionated potentials, primarily due to loss of complex fractionated potentials, as also observed in the decreasing FDT. In comparison to the impact of low-pass filtering however, this effect is clearly less substantial. On the contrary, notch filtering increased the presence of CFPs (max of 1.0017% increase) by adding artificial components to the unipolar fibrillation potential, just as in bipolar measurements [3].
Our results also stress the significance of adequate (i.e., as minimal as possible) filtering for voltage mapping, as unipolar potential amplitude was impacted by high-pass, low-pass, and notch filtering. Low-voltage areas (< 0.5 mV) have been linked to fibrosis, poor cell-to-cell coupling, slowed and discontinuous electrical conduction, and thus maintenance of AF, motivating targeted ablation of these areas as an isolated approach or in addition to CFAE ablation and/or PV isolation [26,27]. As our results indicate, low-pass filtering decreased the number of lowamplitude deflections, thereby reducing the number of potential target sites for low-voltage ablation in clinical practice. On the contrary, high-pass filtering attenuated the overall deflection amplitude and induced an increase in the number of low-amplitude signals, potentially leading to erroneous overdetection of lowvoltage target sites. Notch filtering did significantly lower deflection amplitude, but its impact is rather small (− 0.0163-0.0000 mV change in amplitude).
With increasing filtering (i.e., increasing high-pass or lowering low-pass cut-off frequencies), the number of detected fibrillation potentials declined. In addition, as described in the Supplemental material, filtering impacted timing of LAT. Though not specifically analyzed in this study, the missing potentials as well as the shift in LATs could lead to changes in detected activation patterns during AF. As such, inadequate filtering could result in unintentional under-or overestimation of rotational activity and focal, or peripheral, waves during mapping. A future study on the precise impact of filtering on measured activation sequences during AF could be very insightful.
As implied in the word itself, filtering inherently results in loss of information. Nevertheless, filtering does not have to be inaccurate if the lost information is irrelevant to your case. The challenge always lies in balancing minimal filtering with maximal signal quality (e.g., without artifacts and noise). However, since the relevant frequency content of fractionated fibrillation potentials is unknown and physiological discrimination between true fractionation and noise contribution is (yet) unfeasible, this study validates the use of minimal filtering. A safer alternative to filtering could be implementation of a signal-to-noise ratio (SNR) within mapping systems, in which detection criteria for fractionated and/or low-voltage potentials become stricter in case of considerable noise, without affecting original signal morphology. Using such an approach also prevents artifacts induced by filtering, such as ringing artifacts, to manifest.
Due to its ability to reduce far-field potentials by subtracting two unipolar electrograms at adjacent sites, bipolar electrograms are clinically often preferred above unipolar measurements. [11] For purposes of fractionation analyses, this favor is perhaps undeserved, since apart from filtering bipolar electrogram morphology is also susceptible to changes in interelectrode distance, electrode size and wavefront direction [3,11,12,28] The Pathophysiology of Fractionation The search for true-pathologic-fractionation remains an ongoing challenge. Differentiating between physiologic fractionation, due to, e.g., overlaying myocardial fibers and functional anisotropy, and pathologic fractionation for now remains difficult. The fact that measurement settings, such as filtering, affect fractionation as well provides another challenge. Nevertheless, the potential of fractionation-guided ablation could be significant if one could find methods for discerning true fractionation identifying abnormal conduction and arrhythmogenicity in AF patients. Specifically highresolution AF mapping studies, combined with imaging and/ or histologic techniques, could aid in unraveling the "fact or artifact" of fractionation.

Limitations
In this observational study, we included ten patients in whom we measured the impact of high-pass, low-pass, and notch filtering. Even though sample size seems rather small, a total of 3000 s of AF recordings, consisting of 2,557,045 fibrillation potentials were analyzed. Furthermore, our results indicate the impact of filtering to be very much alike between patients. Considering that filtering is a highly reproducible technique, we hypothesize that the general conclusions of this study can be extrapolated to each individual patient. Nevertheless, in clinical practice, there may be variations in filter properties. For example, to assure generalizability and reproducibility we used zero-phase filtering, but this might not be possible in daily clinical practice. For this study, we used unipolar intra-operative high-resolution mapping data, which is different from (bipolar) endocardial mapping data during ablative therapy with typically lower resolution data. For establishing the impact of filtering on unipolar data however, not spatial resolution but temporal resolution-thus sampling rate-is an important and relevant property, since filtering is done in time-domain. Clinically used mapping systems have comparable sampling rates of~1 or 1.2 kHz, so our results should be considered relevant for endocardial unipolar mapping as well. Increasing sampling rate and resolution of analog to digital conversion and using wider band-pass filtering of origin could potentially further improve signal quality.

Conclusions
High-pass, low-pass, and notch filtering impacted morphology of unipolar fibrillation potentials, including amplitude, fractionation, and FDT, and decreased the number of detected fibrillation potentials, becoming a potential source of error in identification of low-voltage areas and (complex) fractionated potentials. While searching for ablative targets during clinical mapping, operators should be well aware of the consequences of filtering. In case of considerable noise, application of a signal-to-noise ratio-not affecting original signal morphology-could be a safer alternative.