Detection of fractional signals in stable arrhythmias

Abstract

In one example, a method includes receiving a cardiac signal sensed by electrodes at a tissue location within the heart. A fraction is identified in the cardiac signal. The fraction identified at the tissue location is compared between a first cardiac cycle and a second cardiac cycle of the cardiac signal. Based on the comparison, a likelihood that the tissue location is causing a stable arrhythmia is estimated. Based on the estimated likelihood, the tissue location is indicated to a user as likely causing a stable arrhythmia.

Description

[Technical field]

[0001] The present disclosure relates generally to the analysis of electrophysiological (EP) signals, and more particularly to methods for detecting fractionated electrograms in stable arrhythmias. [Background technology]

[0002] Isolating complex fractionated atrial electrograms (CFAEs) to characterize arrhythmias has been previously suggested in the literature. For example, the conference paper "A new approach for automated location of active segments in intracardiac electrograms," World Congress on Medical Physics and Biomedical Engineering, September 7-12, 2009, Munich, Germany, Nguyen MP et al., describes a method for localizing signal complexes corresponding to electrophysiological activity. The disclosed method identifies CFAEs by applying a nonlinear energy operator (NLEO) followed by Gaussian smoothing to time-discrete atrial EGM signals. The method can thus identify regions in the atrial tissue with complex fractionated atrial electrograms (CFAEs) that are specifically involved in the maintenance of atrial fibrillation (AFib). These regions are ideal target sites for ablation to eliminate AFib and restore normal rhythm. The method introduces automated identification of CFAEs using signal processing algorithms to aid in the development of objective strategies for AFib ablation. The idea behind this algorithm is based on that of the Pan-Tompkins QRS detection algorithm. However, in this approach, the signal feature extracted is the signal energy, and therefore the algorithm takes into account not only frequency information but also amplitude information. Adaptive thresholding allows the algorithm to manage changes in signal dynamics. The results have been validated by experts and the algorithm shows a robust performance.

[0003] As another example, Frontera A et al., "Electrogram signature of specific activation patterns: Analysis of atrial tachycardias at high-density endocardial mapping", Heart Rhythm. 2018 Jan; 15(1): Figs. 28-37, describe results from 25 consecutive patients who underwent high-density atrial mapping during atrial tachycardia. Bipolar EGMs were recorded with a 64-pole basket catheter. The following atrial phenomena were identified: slow conduction (SC) areas, line of block (LB), wavefront collision (WFC), pivot site (PS), and gap. EGMs collected at these predefined areas were analyzed for amplitude, duration, and morphology. It was found that specific EGM characteristics in atrial tachycardias could be reproducibly associated to electrophysiological mechanisms. High voltage and short duration EGMs are associated with collision sites and PS that are unlikely to form critical sites for ablation. Long duration low voltage EGMs are associated with SC. However, not all SC regions are within the critical circuit, and identification by EGM characteristics alone cannot guide ablation. [Brief description of the drawings]

[0004] A more complete understanding of the present disclosure will be obtained from the following detailed description of the embodiments of the present disclosure when read in conjunction with the drawings. [Figure 1] FIG. 1 is a schematic, pictorial illustration of a catheter-based electrophysiological (EP) mapping system, according to an example of the present disclosure. [Diagram 2] 1 is a flow chart that generally illustrates a method and algorithm for detection of fractional signals in stable arrhythmias, according to an example of the present disclosure. [Diagram 3] 2 shows an exemplary graph of an electrogram acquired using the system of FIG. 1 , where the electrogram is superimposed with an expanded window of interest (WOI) in the steps described in FIG. 2 , according to one example of the present disclosure. [Figure 4]3 is an exemplary graph of an electrogram acquired using the system of FIG. 1 and the signal after applying the NLEO and Gaussian filter steps described in FIG. 2 according to one example of the present disclosure. [Diagram 5] 3 is a graph that illustrates a schematic representation of two fractions with respective windows found in the steps described in FIG. 2 and the intersection area of ​​the windows, according to an example of the present disclosure. [Figure 6] 3 is a graph that illustrates a schematic representation of the fraction duration fine-tuning step described in FIG. 2 according to an example of the present disclosure. [Figure 7A] 1 is a diagram of two types of signals, the top line being a schematic graph of fractionation and the bottom line being a schematic graph of noise, according to an example of the present disclosure; [Figure 7B] 1 is a diagram of two types of signals, the top line being a schematic graph of fractionation and the bottom line being a schematic graph of noise, according to an example of the present disclosure; [Figure 8] 1 is a schematic volume rendering of an EP map of a ventricle showing tissue locations and fractional magnitudes that may cause a stable arrhythmia, according to an example of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0005] overview Stable arrhythmias, such as atrial flutter or stable ventricular tachycardia (VT), are often defined as sustained arrhythmias that are not associated with significant hemodynamic impairment. However, stable arrhythmias can easily deteriorate into life-threatening conditions and therefore, once diagnosed, require prompt treatment, including the option of ablation of the arrhythmogenic tissue locus causing the stable arrhythmia.

[0006] To characterize a patient's stable arrhythmias, a catheter-based electrophysiological (EP) mapping system may be used to generate an EP map of at least a portion of the patient's heart, such as an EP map of a ventricle. In a typical catheter-based EP mapping procedure, a distal end of a catheter with one or more sensing electrodes is inserted into a heart chamber to sense EP signals, such as unipolar and / or bipolar electrograms (EGMs). As a physician operating the system moves the distal end inside the ventricle, the EP mapping system acquires EP signals at various locations on the inner surface of the ventricle, as well as the respective locations of the distal end. Based on these acquired signals, a processor of the mapping system generates the requisite EP map, including local EP tissue characteristics (e.g., local cycle times between activations) overlaid on an anatomical map of the ventricle.

[0007] When mapping stable arrhythmias (e.g., stable tachycardia), physicians look for specific regions on the surface of cardiac tissue with abnormal EP properties, i.e., one or more tissue isthmuses, to ablate. Typically, isthmuses are regions of EP conduction between adjacent scars or between a natural structural barrier and a scar, because such regions facilitate stable arrhythmias. Isthmuses are characterized by EP signals with fractional activation.

[0008] Current systems using current algorithms for identifying and characterizing stable arrhythmias do not provide the physician with a good pointer to the area to ablate, and therefore the physician must typically perform time-consuming checks of candidate tissue locations prior to ablation.

[0009] Fractionated electrograms (including irregular patterns such as bursts of very rapid deflection of EGMs) can be classified into two main types. Some fractionated electrograms do not exhibit characteristic cycle times. Other fractionated electrograms exhibit approximately characteristic cycle times between episodes of signal fractionation.

[0010] Examples of the disclosure described herein provide improved methods and systems for mapping cardiac tissue locations that exhibit approximately periodic cycle times characteristic of stable arrhythmias, such as left atypical atrial flutter. The disclosed techniques identify electrograms that exhibit repeatable fractional signals and use these electrograms to map cardiac tissue locations that are likely causing stable arrhythmias.

[0011] Using the disclosed techniques, a physician is presented with an EP map overlaid with tissue locations that are likely to cause stable arrhythmias. The likelihood metric may be configurable by the physician. Given the disclosed presentation, a physician can easily and interactively evaluate the map to determine which cardiac tissue locations are indicative of stable arrhythmias and whether to ablate tissue there.

[0012] Thus, some examples of the present disclosure described herein provide methods and systems for automatically identifying and analyzing cardiac tissue regions that cause stable arrhythmias by identifying and analyzing electrograms indicative of recurrent fractions.

[0013] In some examples, the processor can analyze intracardiac electrogram (EGM) and / or surface electrocardiogram (ECG) signals to calculate arrhythmia cycle length and then identify repeatable fractional activations in the EGM on a beat-by-beat basis. These activations are not exactly repeatable, but are repeatable enough for the disclosed algorithms to identify such repeatable fractions by comparing the current beat (i.e., cardiac cycle) to several previous beats (which may occur a given number of beats prior during the same mapping session).

[0014] In addition to identifying repeatable fractions, the algorithm characterizes the fractions, for example, by finding their duration (in time), ΔT, and their temporal location in the cardiac cycle, T. Both of these quantities are relevant to localizing the isthmus. The longer the fraction, the more likely it is to be indicative of arrhythmogenic tissue. With regard to location, if the timing T of a fraction occurrence occurs in the mid-diastolic region of the cardiac cycle, it is also highly likely to be indicative of arrhythmogenic tissue.

[0015] In some examples, the disclosed algorithms rely on applying a nonlinear energy operator (NLEO) to the input EP signal, which amplifies the fraction, and applying a Gaussian filter to the NLEO result imparts a smoothing envelope effect to the fraction signal.

[0016] The disclosed algorithm includes the following main steps: receiving a cardiac signal sensed by an electrode at a tissue location within the heart; (ii) identifying fractions within the cardiac signal (e.g., across a beat of the signal); (iii) comparing the fraction to a fraction found over a previous cardiac cycle (e.g., between a first cardiac cycle and a second cardiac cycle); (iv) estimating the likelihood that the tissue location is causing a stable arrhythmia based on the comparison (e.g., by estimating the likelihood of a fraction within the signal that is indicative of a stable arrhythmia); and (v) indicating to a user (e.g., with a score) the tissue location as likely to cause a stable arrhythmia based on the estimated likelihood, with a minimum score meaning that the location is unlikely to cause a stable arrhythmia and a maximum score meaning that the location is very likely to cause a stable arrhythmia.

[0017] Further algorithmic steps may include: Calculate the baseline noise of the EP signal and dynamically adjust the threshold for NLEO calculation according to the baseline noise. · Capping of the EP signal before applying the NLEO calculation to improve the results of the NLEO calculation. ·Finding the fraction by using EP signal amplitude along with NLEO. To facilitate identification of shifts in fraction timing, we represent the fraction window as a triangle between 0 and 1. To get a quick measure of similarity between beats, represent the fractionation window as a rectangle between -1 and 1 (speeds up CPU calculations). Requires a minimum similarity between beats. Requires the peak-to-peak of the current and previous beats to be similar to the sanity test. Measuring the "silence ratio" as the duration of no fractions divided by the total cycle time. This ratio may be used as input to a machine learning classifier such as logistic regression, or as part of a sanity test to distinguish repeatable fractions from irrelevant irregular fractions. For example, very long fractions with little silence between fractions are indicative of noise. Use the following as features in a classifier, e.g. a logistic regression classifier: * Duration of the candidate fraction. * The number of candidate fractions per period. * The number of zero crossings per duration. * The logarithm of the number of wavefront candidates (counted by annotations). The calculation of wavefront annotation candidates is taught in US Pat. No. 9,380,953 and US Pat. No. 11,071,489. * Silence ratio that can be capped by a maximum limit. * The "aspect ratio" of the candidate fraction, calculated as the peak-to-peak signal divided by the duration. A higher aspect ratio means it is more likely to be a true fraction (and not noise). A user interface for dynamically selecting the classifier threshold while the user is moving a slider and highlighting the fractional regions on the map according to the slider state in real time. Displaying the fractional area in real time (as the user moves the threshold scoring slider) to allow the user to intuitively select the appropriate balance between sensitivity and specificity.

[0018] For each given EP signal, the algorithm returns a fractionation score. The user sets the fractionation score threshold to balance between sensitivity and specificity. A typical fractionation score can be a number from 0 to 10 that predicts the likelihood of a fractionation component in the signal within a time frame. 0 means fractionation is unlikely to occur, and 10 means fractionation is likely to occur. The score can guide the physician performing EP mapping as to where to look for isthmus tissue.

[0019] The rationale for using a fractional score of 0-10 is to avoid confusion with statistical probability. Similarly, a fractional score of 0-100 can be used. However, a user may understand 100 as a 100% probability of fractionation. Therefore, we believe that a fractional score of 0-10 is a better choice. A fractional score of 10 means that the probability of fractionation is very high, but still does not mean a 100% probability.

[0020] Cardiac conditions with stable arrhythmias associated with repeatable fractions include, but are not limited to, atypical flutter, focal atrial tachycardia (AT), and ischemic VT. Thus, by applying the disclosed method, tissue locations causing many stable arrhythmias can be more accurately identified, leading to safer and more effective treatment of stable arrhythmias by ablation.

[0021] System Description FIGURE 1 is a schematic, pictorial diagram of a catheter-based electrophysiological (EP) mapping system 21, according to one example of the present disclosure. FIGURE 1 illustrates a physician 27 using an electroanatomical mapping catheter 29 to perform electroanatomical mapping of a heart 23 of a patient 25. The mapping catheter 29 includes at its distal end one or more arms 20, each of which is coupled to a bipolar electrode 22 that includes adjacent electrodes 22a and 22b.

[0022] During a mapping procedure, the positions of the electrodes 22 are tracked while they are within the patient's heart 23. For that purpose, electrical signals are passed between the electrodes 22 and external electrodes 24. For example, three external electrodes 24 may be coupled to the patient's chest and another three external electrodes may be coupled to the patient's back (for ease of illustration, only one external electrode is shown in FIG. 1).

[0023] Based on the signals, and taking into account the known locations of the electrodes 24 on the patient's body, the processor 28 calculates an estimated location of each electrode 22 within the patient's heart. Respective electrophysiological data, such as a bipolar electrogram trace, is additionally acquired from the tissue of the heart 23 by using the electrodes 22. The processor may thus associate any given signal received from the electrodes 22, such as a bipolar EP signal, with the location at which the signal was acquired. The processor 28 receives the resulting signals via the electrical interface 35 and uses the information contained in these signals to construct an electrophysiological map 31 and an EGM or ECG trace 40, which are presented on the display 26. The EGM or ECG trace 40 is also stored in the memory 33 for use with the disclosed algorithms. One tracking system and method that may generate the map 31 is an advanced current localization (ACL) system implemented in a variety of medical applications, such as the CARTO™ system manufactured by Biosense-Webster Inc., and is described in detail in U.S. Pat. No. 8,456,182, the disclosure of which is incorporated herein by reference.

[0024] Processor 28 typically comprises a general-purpose computer with software programmed to perform the functions described herein. The software may be downloaded to the computer in electronic form, for example over a network, or alternatively or additionally may be provided and / or stored on a non-transitory tangible medium, such as magnetic, optical, or electronic memory. In particular, processor 28 executes the dedicated algorithms disclosed herein, contained in FIG. 2, which enable processor 28 to perform the steps of the present disclosure, as described further below.

[0025] The exemplary illustration shown in FIG. 1 is chosen solely for conceptual clarity. Other types of electrophysiological sensing catheter geometries may be employed, such as the Lasso® catheter (manufactured by Biosense Webster Inc., Irvine, California). Additionally, a contact sensor may be mounted at the distal end of the mapping catheter 29 to transmit data indicative of the physical quality of the electrode contact with the tissue. In one example, measurements of one or more electrodes 22 may be discarded if their physical contact quality is indicated as poor, while measurements of other electrodes may be considered valid if their contact quality is indicated as sufficient.

[0026] Detection of fractional signals in stable arrhythmias 2 is a flow chart that generally illustrates a method and algorithm for detection of fractional signals in stable arrhythmias, according to one example of the present disclosure. The algorithm executes a process that, according to the example presented, begins with a processor, such as processor 28, receiving (i) a bipolar signal (e.g., a waveform) from catheter 29, and (ii) a respective unipolar signal, at cardiac signal reception step 200. In the example shown, processor 28 receives the unipolar component of the bipolar EGM. After power line interference cancellation, both signals are received and sampled at 1 kHz in millivolts.

[0027] Next, in annotating step 201, processor 28 annotates the activation of the bipolar EGM and each of the aforementioned unipolar EGMs.

[0028] Using the annotations, processor 28 calculates, in a cycle time estimation step 202, the cardiac cycle time (also called cycle length (CL)) in milliseconds as the time difference between the current and previous reference annotations.

[0029] The processor then applies an extended window of interest (WOI), a window of duration greater than the calculated cycle length (typically 120% of the estimated CL), to the bipolar signal in an application WOI step 203. Steps 201-203 allow for isolation of the bipolar waveform for analysis, such as bipolar waveform 301 seen in FIG. 3 below. In particular, using an extended WOI ensures complete capture of all fractions of the signal that start within the cardiac cycle to be analyzed. The rationale for using an extended WOI is that clinically relevant mid-diastolic fractions may exist halfway between two successive WOIs; that is, the fraction starts before the end of one WOI and continues after the start of the next successive WOI. The extended WOI helps detect these fractions by allowing fractions to end after the nominal WOI.

[0030] Next, in a signal filtering step 204, the processor 28 applies high pass and low pass filters (HPF and LPF) to the bipolar signal to further clean the signal.

[0031] Next, in a signal capping step 205, the processor 28 applies capping to the bipolar signal. In this step, the magnitude of the bipolar EGM is capped between a lower cap value and an upper cap value. The input of the capping process is the signal after the HPF and LPF, and the output is the capped EGM. In one example, the upper cap value is calculated as the Nth (e.g., N=10) largest sample of the signal in the extended WOI, and the lower cap value is calculated as the Mth (e.g., M=10) lowest sample of the signal in the extended WOI. The rationale for using the N=M=10th largest and smallest samples is that EGM peaks are typically 10-20 ms wide. Thus, if activation (e.g., far-field activation effects) is present in the signal, excluding 20 samples (the 10 largest and 10 smallest samples) roughly excludes large irrelevant activations that would otherwise distort the subsequent NLEO analysis.

[0032] In the baseline calculation step 206, the processor 28 defines the baseline noise in the WOI as the minimum of a "sliding maximum window" of the absolute value of the signal. The width of the sliding maximum window is a certain percentage of the cardiac cycle length (e.g., 20%), but never narrower than some given duration (e.g., 35 ms). The lower limit of 35 ms has no effect if the cycle length is greater than 175 ms, which means 342 atrial beats / min, since the window size is greater than 35 ms when the cycle length is greater than 175 ms. An example of this scheme for finding the baseline noise segment 305 with the smallest maximum absolute value is shown in FIG. 3. Typically, the same process is repeated for the two previous cardiac cycles, and the largest baseline noise among the last three cardiac cycles is considered as the baseline noise of the signal.

[0033] The rationale for the 20% segment length is that even if ventricular far-field interference is assumed to exist in one of the last three beats, and even if there are very long fractions, there should still be continuous silence for at least 20% of the cardiac cycle time somewhere within the WOI (based on at least a minimum predetermined portion of the cardiac cycle that does not include any fractions). Thus, the maximum absolute value of the signal in the most silent 20% segment is considered as the baseline noise. However, fractions greater than 80% of the cycle length may exist, and in these cases the baseline noise may be overestimated. The baseline noise is used to calculate only the number of fractions and the NLEO threshold constant. Because the disclosed logistic regression classifier uses multiple features, an overestimated baseline noise reduces the likelihood that a fraction will be detected, but even if the baseline noise is overestimated, the fraction may still be detected.

[0034] Next, in a NLEO threshold calculation step 208, the processor 28 uses the baseline noise and EGM capped amplitudes to calculate two thresholds that will be applied to the NLEO results to find a window in which a valid fractionation occurs for further analysis. A first threshold, hereafter referred to as K, is used with a sliding standard deviation operator that is applied to the result of the Gaussian filtered NLEO results. In this first thresholding step, the result of the sliding standard deviation is multiplied by a constant K to obtain a threshold value. The K value is dynamically adjusted in the disclosed method to minimize false positive detection of fractionation. In the disclosed algorithm, the processor dynamically adjusts the K value according to the ratio between the baseline noise and the peak-to-peak value of the EGM signal. In this way, K typically varies between 0.15 for clean signals and 0.35 for noisy signals.

[0035] A second threshold is applied to the amplitude of the Gaussian-filtered NLEO result. Although NLEO is a better predictor of fractionation than just amplitude, amplitude still needs to be taken into account to prevent noise from being detected as fractionation. For this purpose, a sample is considered to be part of a fractional component only if its amplitude is larger than a few μV, typically the resolution of the acquisition hardware, or larger than twice the baseline noise (whichever is larger).

[0036] In an apply NLEO step 210, processor 28 applies the NLEO to the bipolar EGM and then applies a smoothing Gaussian filter to the output of the NLEO, an example of a resulting waveform is shown in Figure 4. As can be seen in Figure 4, the result of step 210 (e.g., waveform 404) amplifies the visibility of fractions in the bipolar EGM 402.

[0037] Next, in find fraction window step 212, processor 28 finds a fraction window, such as window 501 in FIG. 5. The fraction window may be a rectangle in the modulation pattern, with a 0 on the vertical axis meaning no fractionation and a 1 on the vertical axis meaning there is a fractionation at that time. In the next step, to make it easier to align the centers of the fraction window at the previous and current beats, the fraction window is represented as an isosceles triangle. During the remainder of the algorithm, the fraction window is represented as an isosceles triangle. In such a case, a modulation of 0 on the vertical axis means there is no fractionation at that time, and any positive value on the vertical axis means there is a fractionation at that time. As can be seen, the maximum height on the vertical axis is the center of the fraction window.

[0038] Fractionation windows with a gap of less than 30 ms are integrated. The rationale for 30 ms is the following: the absolute refractory period of healthy cardiac cells lasts approximately 180 ms. One of the two conditions required to maintain reentry in cardiac tissue is a short refractory period (the other has slow conduction). Therefore, in the isthmus, the absolute refractory period is expected to be shorter. However, 30 ms is one-sixth of this number. Therefore, if the gap between two fractionation windows is less than 30 ms, it is reasonable to assume that this cannot be a new activation, but is either noise or a continuation of the same activation. In both cases, the activation is integrated. If the second fractionation window is noise, the comparison with the previous beat made in the subsequent step should in any case remove this part of the fraction.

[0039] Beat-to-beat consistency is one of the most important disclosed fractional predictors of successful mapping of stable arrhythmias (such as the map shown in FIG. 8). The rationale is that in stable arrhythmias, with 1:1 atrioventricular conduction, all beats should show similar morphology. With 1:N atrioventricular conduction, all beats with no "overlap of ventricular activity" should be similar as well. Thus, the current beat should be similar to either the previous beat or the beat before that.

[0040] Some heart beats may be masked by ventricular activation far-field signals, so the most similar beat may be the previous beat or the beat before the previous beat.

[0041] The timing of the fractions may also show some shifts between beats. To solve the shift problem, the processor 28 compares the fractions in the window in the current extended WOI with two previous extended WOIs in a fraction comparison step 214. In other words, the processor compares the identified fractions at the tissue location between the first and second cardiac cycles of the cardiac signal. In one example, the processor 28 varies the shift between -20 ms and 20 ms in steps of 2 ms while correlating the fractions. The shift that gives the highest correlation (e.g., dot product) is considered the best shift. However, it is difficult to compare fractions between different beats because although the timing of the fractions in the cycle length is stable enough, the exact form of the fractions tends to vary between beats. To solve this problem, the disclosed technique detects the start and end times of the fraction window, converts the window into a triangle as shown in FIG. 5, and performs the correlation (e.g., dot product) on the triangle converted signal instead of correlating the signal itself. The triangle ensures that the centers of the fraction complexes are aligned as much as possible.

[0042] Once the shift is determined for the two previous beats, the fractionation window as a rectangle is transformed to a scale between -1 and +1 and a dot product is calculated between the current beat and the two previous beats. The result of this dot product is divided by the number of samples in the extended WOI to obtain a measure of similarity between the two beats.

[0043] Transforming the vertical axis to be between -1 and 1 gives the following scheme: Any matching sample contributes positively to similarity, while any non-matching sample imposes a cost on similarity.

[0044] [Table 1]

[0045] If we divide the dot product by the number of samples in the extended WOI, the resulting similarity measure is a real number between -1 and 1.

[0046] Of the two previous beats, the beat with the highest similarity score is considered the most similar beat.

[0047] Next, to verify step 214 , processor 28 performs a “sanity check” in an apply sanity test step 216 .

[0048] A sanity test is performed to determine whether to stop the algorithm and return a fractional score equal to 0 without further processing. If the sanity test fails, it is assumed that there is not enough information to output any non-zero fractional scores. As a design principle, the algorithm prioritizes specificity over sensitivity; in other words, false negatives are preferable to false positives. Thus, if the sanity test fails, the algorithm stops here.

[0049] The sanity test passes if all of the following conditions are met. The constants in the following sections are given as typical examples.

[0050] (1) The peak-to-peak of the most similar previous beat is at least one-third of the peak-to-peak of the current beat over the entire extended WOI. If not, the previous beat is assumed to be just noise.

[0051] (2) The peak-to-peak of the current beat is at least one-third the peak-to-peak of the most similar previous beat in the entire extended WOI. If not, the current beat is assumed to be simply noise.

[0052] (3) The similarity measure computed in the previous step is at least a given fraction (e.g., at least 0.5).

[0053] (4) - The maximum fractional window in any of the previous three beats is at most a maximum fraction of the cycle length (e.g., 80%). If this is not the case, it is assumed that at least some of the fractional window represents noise and the beat is discarded, since the algorithm prefers false negatives over false positives.

[0054] (5) The onset and offset of a potential fraction do not differ by more than a given percentage (e.g., half a fraction) of the duration in the current beat and the most similar previous beat. The sum of the onset and offset differences is at most a given percentage (e.g., 75%) of the fractional duration (after accounting for shifts to obtain best similarity).

[0055] (6) - The duration of the fractional component is at least a given duration (e.g., at least 30 ms).

[0056] The rationale behind Sanity Check 6 (e.g., using a duration of 30 ms) is that the disclosed algorithm is designed to find fractions of at least 30 ms. Narrower fractions are possible, but such fractions are difficult to distinguish from normal activation. Thus, in one example, a lower limit of 30 ms is a given restriction of the algorithm.

[0057] (7)-The duration of a fractional component shall be at most a given percentage (e.g., 85%) of the WOI.

[0058] (8) There shall be at least a minimum number (e.g., 4) of significant extreme values.

[0059] (9) - At least a given percentage (eg, 15%) of the WOI does not include any fraction in the current beat and the most similar previous beat.

[0060] The rationale for the sanity check (e.g., 15%)9 is that the algorithm is designed to find a fraction up to 85% of the WOI in one example. Larger fractions are clinically possible, but such fractions are difficult to distinguish from mere noise. Thus, the upper limit of 85% of the WOI is the limit taken in that example of the algorithm.

[0061] 5, the durations 511 and 513 of windows 501 are different, with the intersection area 505 between the two equal to the smaller duration 511. Thus, to improve detection (e.g., by the classifier) ​​of fractions that pass the sanity check (pass step 216), processor 28 then fine-tunes the onset and offset timing of the fractions, in a fine-tuning step 218.

[0062] Since there may be more than two windows considered for comparison, the intersection duration of the fractional windows of the current beat and the most similar previous beat is calculated. The maximum intersection point starting from the beginning of the WOI and ending at the end of the extended WOI is assumed to be the fractional time frame possible.

[0063] The previous beat is calculated by subtracting one or two cycle lengths from the WOI. The previous WOI is accurate because the cycle length is calculated as the difference between two reference annotations. The WOI before the previous WOI may differ slightly from the corresponding reference annotation. However, the WOI is shifted left or right in any case to find the best alignment of the fractionation window.

[0064] The algorithm is designed to find the fraction up to the end of the extended WOI. In this example, the extended WOI is 20% beyond the end of the original WOI. In some cases, this can cause the fraction to be split into two beats. This is a known limitation of one example of the algorithm.

[0065] Fine-tuning the time frame of possible fractions (i.e., step 218) is done as follows: Because the time frame is calculated based on the intersection between two beats, it may appear to be prematurely trimmed at a beat after the current longer duration, as shown in FIG.

[0066] Thus, processor 28 first shrinks the onset and offset of the time frame to the leftmost and rightmost extreme values ​​by a voltage step of at least the size of the baseline noise, e.g., 3 mV or 5% of the peak-to-peak of the potential fractional component, whichever is greatest.

[0067] The onset and offset are then extended by the processor by up to 20 ms on each side until each side is sufficiently close to zero, which is assumed when the maximum of the absolute values ​​of the next three samples is below three times the baseline noise. This fine-tuning step 218 is shown in FIG.

[0068] The resulting time difference between the onset and offset of each candidate fraction is hereafter also referred to as the "duration" of the fraction or the "time frame" of the fraction.

[0069] In step 220, processor 28 stores the onsets and offsets of the candidate fractions and information about their respective waveforms. Once the time frames of the possible fractions have been calculated, processor 28 continues the algorithm by extracting features of the possible fractions in steps 222-228, which are input for the subsequent classification and scoring steps 230-232.

[0070] In some examples, once candidate fractions have been identified and temporally segmented, the processor extracts fraction features that are used by a regression-based classifier to score fractions between those that are benign and those that are highly likely to represent tissue locations that cause stable arrhythmias.

[0071] The logistic regression classifier used to calculate the signal fraction score is given by:

[0072]

number

[0073] The features fed to the regression classifier are: F1←time frame duration of potential fraction F2 ← number of fractions in the time window of potential fractions divided by duration. F3 ← peak-to-peak voltage of the potential fractional component divided by its duration. F4←natural logarithm of 3+(number of wavefront candidates in the time frame of potential fractions in the current beat or the most similar beat, whichever is larger) F5←Number of zero crossings in the time frame of the potential fraction divided by its duration.

[0074] F1 was already given in step 220.

[0075] In F4 extraction step 222, processor 28 counts the number of wavefront annotation candidates that fall within the time frame of each potential (i.e., candidate) fraction for the current beat as well as the most similar previous beat. The maximum number of annotation candidates between the current beat and the most similar previous beat is assumed to be the actual number of wavefront candidates.

[0076] In the F3 extraction step 224, the processor 28 divides the peak amplitude of the fraction by its relative duration. This feature contributes negatively to the fraction score. A high aspect ratio, as seen in FIG. 7 for the signal in line (a), indicates either a normal signal or noise. A low aspect ratio, as seen in FIG. 7 line (b), indicates a clinically meaningful fraction. The rationale is that clinically meaningful fraction signals, empirically, appear more frequently in low voltage regions. Thus, this feature, which has a negative coefficient in the logistic regression classifier, makes the system more sensitive to fractions in low voltage regions and more specific in high voltage regions.

[0077] In an F2 extraction step 226, processor 28 calculates the number of fractions (potential extrema) within the time window, taking into account only sharp fractions (eg, having an absolute gradient greater than 3 mV between successive samples).

[0078] In an F5 extraction step 228, processor 28 calculates the number of zero crossings within the time frame of the potential fraction, which is defined as the sum of the upward and downward zero crossings. An upper zero crossing is a sample that is 0 or positive with a negative left neighbor. A lower zero crossing is a sample that is 0 or negative with a positive left neighbor.

[0079] training The above logistic regression classifier model has five independent variables (features) and therefore six coefficients are required, including the intercept coefficient k0.

[0080] A good classification result is considered to have an Event Per Variable (EPV) value of at least 20. The disclosed model has been trained on more than 1000 cases, and therefore the EPV value in this training is more than 200. This value is an order of magnitude larger than the proposed value of 20. Therefore, the number of signals used for training can be considered to be more than sufficient.

[0081] Each signal used in the training set was tagged by three clinical account experts and a SW engineer. Only congruently tagged signals were fed to training. The rationale was to not confuse the classifier with ambiguous signals (given specificity is more important than sensitivity).

[0082] Inference (step 232) The trained coefficients are stored as part of the algorithm. These coefficients are used to calculate a fraction score. During a clinical study, a physician may select signals by their fraction score and other attributes. Some of the attributes are duration of potential fraction and peak-to-peak voltage. REF3 demonstrates that duration and voltage of activation are good predictors of slow conduction regions. It is expected that a physician will use clinical judgment and experience to set a threshold value associated with each one of the attributes, including the fraction score, to select signals relevant to the current clinical study.

[0083] Finally, in a map generation step 234, processor 28 generates and displays to the user an EP map such as that shown in Figure 8, overlaid with tissue locations scored according to their estimated influence on stable arrhythmias. Using a scoring scale on the GUI, the user can determine a threshold score, and only tissue locations scored above that threshold will be observed on the map.

[0084] The flowchart shown in Figure 2 has been selected solely for the purposes of conceptual clarity. This embodiment may also include additional steps in the algorithm, such as preselecting the input EGM based on an indication of the degree of physical contact of the electrodes with the diagnosed tissue from a contact force sensor. This step, and other possible steps, have been intentionally omitted from the disclosure herein to provide a more simplified flowchart.

[0085] Figure 3 shows an exemplary graph of an electrogram 301 acquired using the system of Figure 1 according to one example of the present disclosure, where the electrogram 301 has been overlaid with a nominal extended window of interest (WOI) in step (203) described in Figure 2. As can be seen, the disclosed extended WOI captures two candidate fractions 303, since using the nominal WOI would cause truncation and loss of the later (i.e., right) fraction due to falling into the clinically relevant mid-diastolic region of the cardiac cycle.

[0086] FIG. 3 further illustrates a "sliding maximum window" 305 having a width of a given segment of cardiac cycle time (eg, a segment lasting 20%) that is used in calculating the baseline noise of the cardiac cycle in step 206 above.

[0087] 4 shows an example graph of an electrogram 402 acquired using the system of FIG 1 and a signal 404 after applying the NLEO and Gaussian filter steps (step 210 above) described in FIG 2, according to one example of the present disclosure. As can be seen, the NLEO highlights the candidate segments and the signal 404 is slightly smoother than that of the EGM 402.

[0088] FIG. 5 is a graph showing, in accordance with an example of the present disclosure, two fractions with respective windows 501 found in step 212 according to FIG. 2 and the intersection area 505 of the windows. As can be seen, two different windows have durations 511 and 513, the duration 513 of the current beat being longer than the duration 511 of the previous beat. To take into account the durations in the subsequent set, the intersection area is defined between any two windows as the length of the shorter duration. Using the definition, the processor 28 fine-tunes the onset of the offset timing of the fractions in the fine-tuning step 218 described above. Thus, in the case shown, the duration of the intersection area 505 is equal to the duration of the duration 511.

[0089] 5 further illustrates a duration 507 within which there are no fractions in EGM 504. Processor 28 measures the "silence ratio" as the duration 507 within which there are no fractions divided by the total cycle time. This ratio helps distinguish repeatable fractions from irrelevant irregular fractions.

[0090] 6 is a graph that illustrates a schematic of the fine-tuning step of fraction duration (step 218) described in FIG. 2 according to an example of the present disclosure. First, the onset and offset of the time frame are reduced to the left and right extremes that have at least a dV of 3 mV of baseline noise or a dV of 5% of the peak-to-peak of the potential fractional components (whichever is the largest). This step results in a time window 602.

[0091] The onset and offset are then extended by up to 20 ms on each side until they are close enough to 0. This step results in a time window 606. When the maximum of the absolute values ​​of the next three samples is below three times the baseline noise, it is assumed to be close enough to 0.

[0092] 7A and 7B are diagrams of two types of signals with an upper line (FIG. 7A) of a schematic graph of fraction (702, 704, 706) and a lower line (FIG. 7B) of a schematic graph of noise (711, 713, 715), according to one example of the present disclosure.

[0093] Signals 702-706 can be described as "probably fractionated." As can be seen, all three windows of signals 702-706 have relatively high height / width aspect ratios. A physician would probably not mistakenly call them "fractionated."

[0094] On the other hand, all three windows of signals 711-715 have relatively low height / width aspect ratios, and a physician would likely mistakenly consider these artifacts to be true "fractions."

[0095] Empirically, clinically meaningful fractions correlate with lower peak-to-peak voltages. Thus, delivering the peak-to-peak voltage divided by the duration of the potential fractional component will automatically increase the fraction score in low-voltage regions and vice versa.

[0096] To overcome such biases, the disclosed technique introduces an F3 emphasis feature in the form of peak-to-peak voltage of the latent fractional components divided by duration to the classification scoring function above, i.e., to distinguish actual fractions from noise in the scoring.

[0097] FIG. 8 is a schematic volume rendering of an EP map of a ventricle showing tissue locations and fractional magnitudes that may cause stable arrhythmias, according to an example of the present disclosure.

[0098] In the illustrated example, the EP map 800 is a local activation time (LAT) map overlaid with a fraction score graphical display 802 to indicate cardiac tissue locations exhibiting fractions indicative of stable arrhythmias.

[0099] As discussed above, the physician can adjust the scale 804 in the GUI to show only those tissue locations that have a score above a threshold (in the illustrated case, those having a score >7).

[0100] Based on EP map 800, a physician, such as physician 27, can plan and perform careful selective ablations to eliminate stable arrhythmias while minimizing risk to the patient. EXAMPLES

[0101] The method includes receiving a cardiac signal sensed by electrodes (22) at a tissue location within the heart. A fraction (303) is identified in the cardiac signal. The identified fraction at the tissue location is compared between a first cardiac cycle and a second cardiac cycle of the cardiac signal. Based on the comparison, a likelihood that the tissue location is causing a stable arrhythmia is estimated. Based on the estimated likelihood, the tissue location is indicated to a user as likely causing a stable arrhythmia. EXAMPLES

[0102] The system of example 1, wherein indicating the tissue location includes indicating to the user a score of the likelihood that the tissue location is causing a stable arrhythmia. EXAMPLES

[0103] The system of example 1 or example 2, wherein indicating the tissue location includes recommending ablation of the tissue location. EXAMPLES

[0104] 2. The system of example 1, wherein receiving a cardiac signal includes receiving a bipolar electrogram. EXAMPLES

[0105] 2. The method of example 1, wherein identifying the fraction comprises estimating a baseline noise of the cardiac signal. EXAMPLES

[0106] Identifying the fraction comprises dynamically adjusting the threshold of a nonlinear energy operator (NLEO) according to the ratio between the baseline noise and the peak-to-peak value of the EGM signal, where the baseline noise is given by the minimum value of a sliding maximum window of the absolute value of the signal, as described in Example 5. EXAMPLES

[0107] 2. The method of example 1, wherein identifying the fraction comprises comparing a period of the cardiac signal during which at least a minimum predetermined portion of the cardiac cycle does not include any fraction. EXAMPLES

[0108] 2. The method of example 1, wherein identifying the fraction comprises calculating a ratio between a peak-to-peak voltage of the cardiac signal over a time interval and a duration of the time interval, and using the ratio to distinguish the fraction from noise. EXAMPLES

[0109] The method of example 1, wherein estimating the likelihood comprises applying a logistic regression-based classifier. EXAMPLES

[0110] The method of Example 1, wherein comparing the fractions includes adjusting a time onset and a time offset of at least one of the fractions. EXAMPLES

[0111] The method of example 1, wherein identifying the fractions in a given cardiac cycle includes capturing all fractions that begin within the cardiac cycle using an extended window of interest (WOI). EXAMPLES

[0112] The method of example 1, wherein estimating the likelihood is performed based on the number of zero crossings occurring in the cardiac signal within each fraction. EXAMPLES

[0113] 2. The method of embodiment 1, wherein estimating the likelihood is performed based on the number of local extrema occurring in the low-pass filtered cardiac signal within each fraction. EXAMPLES

[0114] The method of example 1, wherein comparing the fractions includes representing the fraction windows as triangles. EXAMPLES

[0115] 2. The method of claim 1, wherein indicating the tissue location includes generating and displaying to the user an electrophysiological (EP) map (800) overlaid with a score of the likelihood that the tissue location is causing a stable arrhythmia. EXAMPLES

[0116] 17. The method of claim 16, comprising adjusting a threshold score using a scale (804) on a graphical user interface (GUI), and only tissue locations whose scores exceed the threshold are displayed on the EP map (800). EXAMPLES

[0117] A system including a memory (33) and a processor (28), wherein the memory (33) is configured to store a cardiac signal, and wherein the processor (28) is configured to (i) receive a cardiac signal sensed by the electrodes (22) at a tissue location within the heart (23), (ii) identify fractions (303) in the cardiac signal, (iii) compare the identified fractions at the tissue location between first and second cardiac cycles of the cardiac signal, (iv) estimate a likelihood that the tissue location is causing a stable arrhythmia based on the comparison, and (v) indicate to a user that the tissue location is likely causing a stable arrhythmia based on the estimated likelihood. EXAMPLES

[0118] The system of Example 15, wherein the processor (28) is configured to indicate the tissue location by presenting to the user a score of the likelihood that the tissue location is causing a stable arrhythmia. EXAMPLES

[0119] The system of Example 15, wherein the processor (28) is configured to indicate the tissue location by generating and displaying an electrophysiological (EP) map (800) to a user, overlaid with a score of the likelihood that the tissue location is causing a stable arrhythmia. EXAMPLES

[0120] 17. The system of claim 16, further comprising a scale (804) on a graphical user interface (GUI) that a user can use to adjust a threshold score, and only tissue locations whose scores exceed the threshold are displayed on the EP map (800).

[0121] Although the embodiments described herein primarily address cardiac diagnostic applications, the methods and systems described herein may also be used in other medical applications.

[0122] It will be understood that the embodiments described above are given by way of example, and that the present disclosure is not limited to what has been particularly shown and described hereinabove, but rather the scope of the present disclosure includes both combinations and subcombinations of the various features described in the above specification, as well as variations and modifications thereof not disclosed in the prior art that would occur to one skilled in the art upon reading the above description.

[0123] [Embodiment] (1) A system comprising: a memory configured to store the cardiac signal; 1. A processor comprising: receiving cardiac signals sensed by electrodes at tissue locations within the heart; Identifying segments within the cardiac signal; comparing the identified fraction at the tissue location between a first cardiac cycle and a second cardiac cycle of the cardiac signal; estimating a likelihood that the tissue location is causing a stable arrhythmia based on the comparison; and a processor configured to indicate to a user the tissue location as likely causing the stable arrhythmia based on the estimated likelihood. (2) The system of embodiment 1, wherein the processor is configured to indicate the tissue location by indicating to the user a score of the likelihood that the tissue location is causing the stable arrhythmia. (3) The system of embodiment 1, wherein the processor is configured to indicate the tissue location by recommending ablation of the tissue location. (4) The system of embodiment 1, wherein the processor is configured to receive the cardiac signal by receiving a bipolar electrogram. (5) The system of embodiment 1, wherein the processor is configured to identify the fraction by estimating baseline noise of the cardiac signal.

[0124] (6) The system of embodiment 5, wherein the processor is configured to identify the fraction by dynamically adjusting the threshold for a nonlinear energy operator (NLEO) according to a ratio between a baseline noise and a peak-to-peak value of the EGM signal, the baseline noise being given by a minimum value of a sliding maximum window of the absolute value of the signal. (7) The system of embodiment 1, wherein the processor is configured to identify the fraction by comparing periods of the cardiac signal during which at least a minimum predetermined portion of the cardiac cycle does not include any fractions. (8) The system of embodiment 1, wherein the processor is configured to identify the fraction by calculating a ratio between a peak-to-peak voltage of the cardiac signal over a time interval and a duration of the time interval and using the ratio to distinguish the fraction from noise. (9) The system of embodiment 1, wherein the processor is configured to estimate the likelihood by applying a logistic regression-based classifier. (10) The system of embodiment 1, wherein the processor is configured to compare the fractions by adjusting a time onset and a time offset of at least one of the fractions.

[0125] (11) The system of embodiment 1, wherein the processor is configured to identify the fractions in a given cardiac cycle by capturing all fractions starting within the cardiac cycle using an extended window of interest (WOI). (12) The system of embodiment 1, wherein the processor is configured to estimate the likelihood based on a number of zero crossings occurring in the cardiac signal within each fraction. (13) The system of embodiment 1, wherein the processor is configured to estimate the likelihood based on the number of local extrema occurring in the low-pass filtered cardiac signal within each fraction. (14) The system of embodiment 1, wherein the processor is configured to compare the fractions by representing fraction windows as triangles. (15) The system of embodiment 1, wherein the processor is configured to indicate the tissue location by generating and displaying to the user an electrophysiological (EP) map overlaid with a score of the likelihood that the tissue location is causing the stable arrhythmia.

[0126] (16) The system of embodiment 15, further comprising a scale on a graphical user interface (GUI) that the user can use to adjust a threshold score, and only tissue locations whose scores exceed the threshold are observed on the EP map. (17) A method comprising the steps of: receiving a cardiac signal sensed by an electrode at a tissue location within the heart; identifying a segment in the cardiac signal; comparing the identified fraction at the tissue location between a first cardiac cycle and a second cardiac cycle of the cardiac signal; estimating a likelihood that the tissue location causes a stable arrhythmia based on the comparison; and and indicating to a user the tissue location as likely causing the stable arrhythmia based on the estimated likelihood. (18) The method of embodiment 17, wherein indicating the tissue location includes indicating to the user a score of the likelihood that the tissue location is causing the stable arrhythmia. (19) The method of embodiment 17, wherein indicating the tissue location includes recommending ablation of the tissue location. (20) The method of embodiment 17, wherein receiving the cardiac signal includes receiving a bipolar electrogram.

[0127] (21) The method of embodiment 17, wherein identifying the fraction includes estimating a baseline noise of the cardiac signal. (22) The method of embodiment 21, wherein identifying the fraction comprises dynamically adjusting the threshold for a nonlinear energy operator (NLEO) according to a ratio between a baseline noise and a peak-to-peak value of the EGM signal, the baseline noise being given by a minimum value of a sliding maximum window of the absolute value of the signal. (23) The method of embodiment 17, wherein identifying the fractions includes comparing periods of the cardiac signal during which at least a minimum predetermined portion of the cardiac cycle does not include any fractions. (24) The method of embodiment 17, wherein identifying the fraction includes calculating a ratio between a peak-to-peak voltage of the cardiac signal over a time interval and a duration of the time interval, and using the ratio to distinguish the fraction from noise. (25) The method of embodiment 17, wherein estimating the likelihood comprises applying a logistic regression-based classifier.

[0128] (26) The method of embodiment 17, wherein comparing the fractions includes adjusting a time onset and a time offset of at least one of the fractions. (27) The method of embodiment 17, wherein identifying the fractions in a given cardiac cycle includes capturing all fractions beginning within the cardiac cycle using an extended window of interest (WOI). (28) The method of embodiment 17, wherein estimating the likelihood is performed based on the number of zero crossings occurring in the cardiac signal within each fraction. (29) The method of embodiment 17, wherein estimating the likelihood is performed based on the number of local extrema occurring in the low-pass filtered cardiac signal within each fraction. (30) The method of embodiment 17, wherein comparing the fractions includes representing the fraction windows as triangles.

[0129] (31) The method of embodiment 17, wherein indicating the tissue location includes generating and displaying to the user an electrophysiological (EP) map overlaid with a score of the likelihood that the tissue location is causing the stable arrhythmia. (32) The method of embodiment 31, comprising adjusting a threshold score using a scale on a graphical user interface (GUI), wherein only tissue locations whose scores exceed the threshold are displayed on the EP map.

Claims

1. 1. A system comprising: a memory configured to store the cardiac signal; 1. A processor comprising: receiving cardiac signals sensed by electrodes at tissue locations within the heart; Identifying segments within the cardiac signal; comparing the identified fraction at the tissue location between a first cardiac cycle and a second cardiac cycle of the cardiac signal; estimating a likelihood that the tissue location causes a stable arrhythmia based on the comparison; and a processor configured to indicate to a user the tissue location as likely causing the stable arrhythmia based on the estimated likelihood.

2. The system of claim 1 , wherein the processor is configured to indicate the tissue location by indicating to the user a score of the likelihood that the tissue location is causing the stable arrhythmia.

3. The system of claim 1 , wherein the processor is configured to indicate the tissue location by recommending ablation of the tissue location.

4. The system of claim 1 , wherein the processor is configured to receive the cardiac signal by receiving a bipolar electrogram.

5. The system of claim 1 , wherein the processor is configured to identify the fraction by estimating a baseline noise of the cardiac signal.

6. 6. The system of claim 5, wherein the processor is configured to identify the fraction by dynamically adjusting the threshold for a nonlinear energy operator (NLEO) according to a ratio between a baseline noise and a peak-to-peak value of the EGM signal, the baseline noise being given by a minimum value of a sliding maximum window of an absolute value of the signal.

7. The system of claim 1 , wherein the processor is configured to identify the fractions by comparing periods of the cardiac signal during which at least a minimum predetermined portion of the cardiac cycle does not include any fractions.

8. 2. The system of claim 1, wherein the processor is configured to identify the fraction by calculating a ratio between a peak-to-peak voltage of the cardiac signal over a time interval and a duration of the time interval and using the ratio to distinguish the fraction from noise.

9. The system of claim 1 , wherein the processor is configured to estimate the likelihood by applying a logistic regression based classifier.

10. The system of claim 1 , wherein the processor is configured to compare the fractions by adjusting a time onset and a time offset of at least one of the fractions.

11. 2. The system of claim 1, wherein the processor is configured to identify the fractions in a given cardiac cycle by capturing all fractions starting within the cardiac cycle using an extended window of interest (WOI).

12. The system of claim 1 , wherein the processor is configured to estimate the likelihood based on a number of zero crossings occurring in the cardiac signal within each fraction.

13. The system of claim 1 , wherein the processor is configured to estimate the likelihood based on a number of local extrema occurring in a low-pass filtered cardiac signal within each fraction.

14. The system of claim 1 , wherein the processor is configured to compare the fractions by representing fraction windows as triangles.

15. 2. The system of claim 1, wherein the processor is configured to indicate the tissue location by generating and displaying to the user an electrophysiological (EP) map overlaid with a score of the likelihood that the tissue location is causing the stable arrhythmia.

16. The system of claim 15, further comprising a scale on a graphical user interface (GUI) that the user can use to adjust a threshold score, and only tissue locations whose scores exceed the threshold are observed on the EP map.

17. 1. A method comprising: receiving a cardiac signal sensed by an electrode at a tissue location within the heart; identifying a segment in the cardiac signal; comparing the identified fraction at the tissue location between a first cardiac cycle and a second cardiac cycle of the cardiac signal; estimating a likelihood that the tissue location causes a stable arrhythmia based on the comparison; and and indicating to a user the tissue location as likely causing the stable arrhythmia based on the estimated likelihood.

18. 20. The method of claim 17, wherein indicating the tissue location includes indicating to the user a score of the likelihood that the tissue location is causing the stable arrhythmia.

19. The method of claim 17 , wherein indicating the tissue location includes recommending ablation of the tissue location.

20. 20. The method of claim 17, wherein receiving the cardiac signal comprises receiving a bipolar electrogram.

21. The method of claim 17 , wherein identifying the fraction comprises estimating a baseline noise of the cardiac signal.

22. 22. The method of claim 21, wherein identifying the fraction comprises dynamically adjusting the threshold for a nonlinear energy operator (NLEO) according to a ratio between a baseline noise and a peak-to-peak value of the EGM signal, the baseline noise being given by a minimum value of a sliding maximum window of the absolute value of the signal.

23. 20. The method of claim 17, wherein identifying the fractions comprises comparing periods of the cardiac signal during which at least a minimum predetermined portion of the cardiac cycle does not include any fractions.

24. 18. The method of claim 17, wherein identifying the fraction comprises calculating a ratio between a peak-to-peak voltage of the cardiac signal over a time interval and a duration of the time interval, and using the ratio to distinguish the fraction from noise.

25. The method of claim 17 , wherein estimating the likelihood comprises applying a logistic regression based classifier.

26. 20. The method of claim 17, wherein comparing the fractions includes adjusting a time onset and a time offset of at least one of the fractions.

27. 20. The method of claim 17, wherein identifying the fractions in a given cardiac cycle includes using an extended window of interest (WOI) to capture all fractions beginning within the cardiac cycle.

28. The method of claim 17 , wherein estimating the likelihood is performed based on a number of zero crossings occurring in the cardiac signal within each fraction.

29. The method of claim 17 , wherein estimating the likelihood is performed based on the number of local extrema occurring in the low-pass filtered cardiac signal within each fraction.

30. The method of claim 17 , wherein comparing the fractions includes representing the fraction windows as triangles.

31. 20. The method of claim 17, wherein indicating the tissue location includes generating and displaying to the user an electrophysiological (EP) map overlaid with a score of the likelihood that the tissue location is causing the stable arrhythmia.

32. 32. The method of claim 31, comprising adjusting a threshold score using a scale on a graphical user interface (GUI), such that only tissue locations with scores above the threshold are displayed on the EP map.

Images