Noninvasive localization of cardiac arrhythmias using electromechanical wave imaging

Abstract

Cardiac arrhythmias are a major cause of morbidity and mortality worldwide. The 12-lead electrocardiogram (ECG) is the current noninvasive clinical tool used to diagnose and localize cardiac arrhythmias. However, it has limited accuracy and is subject to operator bias. Here, we present electromechanical wave imaging (EWI), a high frame rate ultrasound technique which can noninvasively map with high accuracy the electromechanical activation of atrial and ventricular arrhythmias in adult patients. This study evaluates the accuracy of EWI for localization of various arrhythmias in all four chambers of the heart prior to catheter ablation. Fifty-five patients with an accessory pathway (AP) with Wolff-Parkinson-White syndrome (WPW), premature ventricular complexes (PVC), atrial tachycardia (AT), or atrial flutter (AFL) underwent transthoracic EWI and 12-lead ECG. 3-dimensional (3D) rendered EWI isochrones and 12-lead ECG predictions by six electrophysiologists were applied to a standardized segmented cardiac model, and subsequently compared to the region of successful ablation on 3D electroanatomical maps generated by invasive catheter mapping. There was significant inter-observer variability amongst 12-lead ECG reads by expert electrophysiologists. EWI correctly predicted 96% of arrhythmia locations as compared with 71% for 12-lead ECG analyses [unadjusted for arrhythmia type: odds ratio (OR): 11.8; 95% confidence interval (CI): 2.2-63.2; P = 0.004; adjusted for arrhythmia type: OR: 12.1; 95% CI: 2.3-63.2; P = 0.003]. This double-blinded clinical study demonstrates that EWI can localize atrial and ventricular arrhythmias including WPW, PVC, AT, and AFL. EWI when used with ECG may allow for improved treatment for patients with arrhythmias.

One sentence summary:

Electromechanical wave imaging accurately localizes atrial and ventricular cardiac arrhythmias in adult patients in a double-blinded clinical study.

Introduction:

Cardiac arrhythmias (irregular heartbeats) are a cause of morbidity and mortality, often necessitating invasive catheter ablation for curative treatment (1). Diagnosis and localization of arrhythmias are critical for clinical decision making and treatment planning. During invasive electrophysiology study (EPS), extensive operating room time may be required to build detailed anatomical activation maps of the heart during the arrhythmia. Arrhythmias on the left side of the heart may necessitate trans-septal puncture and therapeutic systemic anticoagulation to prevent clot formation and arterial embolism. The current standard of care for diagnosis and localization is clinical interpretation of the 12-lead ECG (2–9). Herein, we study the use of electromechanical wave imaging (EWI), a noninvasive, ultrasound-based imaging modality for localization of cardiac arrhythmias such as Wolff-Parkinson-White (WPW), premature ventricular complexes (PVC), atrial tachycardia (AT), and atrial flutter (AFL).

Electrocardiogram (ECG) algorithms have been proposed to aid in localization of arrhythmias but have varying accuracy, and inter-observer variability is common. Studies of algorithms for localizing accessory pathways and focal ATs have accuracies of 90-93% (10–13), whereas localization of ectopic ventricular rhythms is between 72-82% (14–16). However, the accuracy of these ECG algorithms is lower in real world clinical practice. EWI, a readily available and portable, noninvasive, and non-ionizing imaging modality with the ability to localize arrhythmias, has the potential to facilitate pre-procedural discussions with patients, preoperative planning, and to reduce procedural catheter mapping times.

EWI is a high frame rate ultrasound technique that can map the electromechanical wave corresponding to the propagation of the contraction onset in response to the heart’s electrical activation (17). Unlike mechanical strain-based techniques such as tissue Doppler imaging (TDI) and speckle tracking echocardiography (STE), EWI relies on incremental axial strains. EWI detects small local contractions on the order of 0.01% and tracks inter-frame axial displacement of about 0.01 mm, whereas TDI uses peak systolic longitudinal strain or global regional accumulated longitudinal strains of about 30% throughout systole (18). In addition, TDI is an angle-dependent technique, whereas EWI activation maps are angle independent (19). STE estimates displacements in any direction by tracking the frame-to-frame movement of speckle patterns on 2-dimensional (2D) B-mode images at 100-fold lower temporal resolution than EWI, and is less accurate than radiofrequency-based (RF) cross-correlation performed in the time domain (20). The challenge for producing accurate electromechanical activation maps is the brief length of the electrical activation. Because ventricular depolarization occurs within 50-60 milliseconds, mapping requires a resolution of a few milliseconds (ms) (21). Therefore, high frame rate ultrasound sequences, in this case up to 2000 Hz, are essential (22). Furthermore, EWI processing relies on the 1-dimensional RF signals for high-precision time-domain displacement estimation and does not use conventional B-mode speckletracking on 2D images (23).

Our group has made advances in EWI over the past 10 years using large animal models and echocardiograms on healthy humans to demonstrate that electromechanical activation is correlated with the electrical activation sequence in the heart (24, 25). EWI isochrones have been shown to characterize electromechanical activation patterns of normal sinus rhythm and pacing in a reproducible, view-independent, and angle-independent manner (19, 26, 27). The isochrone generation process does not depend on patient cardiac geometry and there is no anatomical assumption of longitudinal symmetry when rendering the activation maps. EWI has also been used to differentiate epicardial from endocardial ventricular origins in focally-paced canine ventricles (28). Other echocardiography strain-based methods have previously been investigated for identification of arrhythmia such as accessory pathway (AP) localization (29–31). We recently demonstrated clinical feasibility and 100% accuracy of EWI in localizing accessory pathways in minors with WPW (32).

Since EWI is an ultrasound-based technique, it is portable and there is readily available existing infrastructure in hospitals and clinics for potential implementation. In addition, ultrasound is a known cost-effective imaging modality (33). Other noninvasive electrical mapping approaches such as electrocardiographic imaging (ECGI) provide high spatial resolution maps of arrhythmias, however they require computed tomography (CT) or magnetic resonance imaging (MRI), which may be ionizing or time-consuming, to obtain the patient’s cardiac geometry (34–38). ECGI is typically applied to the epicardial surface, however endocardial mapping remains a challenge (39, 40). Finally, there has been some controversy regarding the accuracy of the inverse solution in ECGI (39). Exploration of other noninvasive mapping approaches, such as EWI is therefore warranted.

This clinical study sought to determine the clinical accuracy of transthoracic EWI for noninvasively localizing clinical arrhythmias including WPW, PVC, AT, and AFL in adults using 3D-rendered EWI maps of all four chambers of the heart. We studied a large adult patient population presenting for catheter ablation with pre-existing cardiac disease, previous ablations, and other cardiovascular disease co-morbidities. We compared the diagnostic accuracy of both atrial and ventricular EWI isochrones (multi-2D or 3D-rendered) with 12-lead ECG-based localization by expert electrophysiologists and finally to the gold standard of 3D electroanatomical maps performed with invasive catheter mapping and eventual successful site of ablation.

Results

Spatial resolution of 3D-rendered EWI isochrones

Optimized image acquisition and processing efficiency is paramount for translation of EWI from large animal studies to clinical cases. We investigated the effect of multi-2D sampling on the spatial resolution of 3D-rendered EWI isochrones. We performed two open-chest canine resolution studies to determine the imaging results of an arrhythmic focus located between two of the four standard apical slices. By increasing the number of multi-2D slices sampling the heart’s circumference of interest, the spatial resolution of the earliest activated region increased and the distances computed from the 3D-rendered isochrones between pacing electrodes were closer to the true values measured on the surface of the heart (fig. S1A and fig. S1B). Compared to 4 and 6 multi-2D slices, the 3D-rendered isochrone generated with 12 multi-2D slices displayed early activation with the most distinct focus (area in red) (fig. S1C and fig. S2). In the standard 3D-rendered isochrones generated with 4 multi-2D slices, EWI localized the pacing locations, albeit in a less precise and wider region displayed in orange due to the effect of the interpolation on a broader circumference. We chose to acquire and use 4 multi-2D slices in our clinical study as further increase in number of multi-2D slices acquired would improve the spatial resolution for arrhythmia localization, but would likely prolong EWI scan durations and delay clinical procedure times.

Patients recruited

Sixty-seven patients presenting to the Columbia University cardiac electrophysiology laboratory for catheter ablation of WPW, PVC, AT, and AFL, were consented for EWI. Twelve patients were excluded: 1 withdrew consent prior to imaging, 2 did not demonstrate clinical arrhythmia at the time of EWI scans, and 4 were not imaged for nonstudy related reasons. Five of the sixty (8.3%) imaged patients had poor acoustic windows, preventing thorough echocardiographic visualization and complete acquisition of the required views, which is below standard rates of limited echocardiography studies previously reported at 20% (41). The resulting cohort of fifty-five patients (Fig. 1) underwent EWI (EWI workflow shown in Fig. 2). The mean age was 56.0 ± 2.3 years of age and 71% were men. In this study, 18% (n = 7) of the cases had previously reported wall motion abnormalities detected by conventional 2D echocardiography. Full baseline characteristics are given in Table 1 and other measurements including atrial size, left ventricular ejection fraction, and history of prior ablations or surgeries by subgroup in table S1. Factors that might have limited echocardiography windows and lead to EWI failure for the patient population are listed in table S2. Examples of EWI isochrones in all ventricular and atrial arrhythmias considered are shown in Fig. 3: WPW (Fig. 3A), PVC (Fig. 3B), AT (Fig. 3C), and AFL (Fig. 3D). There were no complications due to EWI within 30 days of the procedures. There were no notable procedural delays due to EWI since scanning took an average of 15 minutes.

Fig. 1. Patient recruitment and study design.

This diagram illustrates patient recruitment, indications for exclusions, and study design.

Fig. 2. EWI workflow.

This figure illustrates the entire EWI processing on the cardiac ventricles of a 26-year-old healthy volunteer. (A) 2D apical views are acquired with a diverging ultrasound sequence at a high frame rate (2000 frames/second). (B) The ventricular myocardium is segmented manually on the low frame rate (30 frames/second) anatomical focused B-mode for each view. (C) Inter-frame displacements and strains are estimated axially on the EWI high frame rate radiofrequency data. The dotted black lines on the ECG represent the onset of the QRS (t=0 for the ventricular four-chamber isochrones) whereas the red dot indicates the frame of interest displayed below. (D) Activation times or zero-crossing locations are selected on the axial inter-frame strain curves. (E) The resulting four 2D electromechanical activation ventricular maps are then co-registered around the left ventricle median axis (longitudinal apex to base rotation axis of the probe, displayed with the dotted black line). Similarly, the four multi-2D atrial maps are generated. Red represents early activation and blue corresponds to late activation (in ms) from the time point of origin on the ECG (p-wave and QRS onset for atrial and ventricular arrhythmia respectively). (F) Finally, the 3D-rendered isochrones are generated for both atria and ventricles by interpolating the multi-2D isochrones around the circumference.

Table 1. Patient characteristics.

Characteristic:	N = 55
Male	39 (71%)
Age (mean)	56.0±2.3 years
Diagnosis
WPW	12 (22%)
PVCs	11 (20%)
AT	7 (13%)
AFL	25 (45%)
Comorbidities
CAD	8 (15%)
CKD	3 (5%)
CHF	18 (33%)
Stroke or TIA	8 (15%)
Atrial Fibrillation	14 (25%)
VT/VF	4 (7%)
Previous Catheter Ablation	7 (13%)
Previous Cardiac Surgery	9 (16%)
On Antiarrhythmic Medication	14 (25%)
Previous echocardiography data	N = 40
LVEF (mean)	45.4±2.6%
LVEF < 50%	17 (42.5%)
LA Diameter (mean)	3.9±0.8 cm
LA Diameter ≥ 4.0 cm	15 (37.5%)
LVEDD (mean)	5.2±0.1 cm
Wall motion abnormality	7 (18%)
History of Limited Echo*	8 (20%)
Body Surface Area (mean)	2.0±0.1 m2
WPW patients	N = 12
QRS duration (mean)	127±3 ms
PVC Patients	N = 11
PVC burden (mean)	29±2%
AFL Patients	N = 25
Cycle Length of AFL (mean)	270±6 ms
AT Patients	N = 7
Cycle Length of focal AT (mean)	362±20 ms

CAD: coronary artery disease; CKD: chronic kidney disease; CHF: congestive heart failure; TIA: transient ischemic attack; VT/VF: ventricular tachycardia / ventricular fibrillation; LA: left atrium; LVEF: left ventricular ejection fraction; LVEDD: left ventricular end diastolic diameter.

^*Previous clinical transthoracic echocardiogram labeled as at least “moderately limited” by the performing sonographer, or one in which a left ventricular EF could not be determined.

Fig. 3. EWI isochrone examples in ventricular and atrial arrhythmias.

This figure shows 3D-rendered EWI isochrones of the four included arrhythmia types. Red represents the earliest activated region and blue the latest. Each case includes the 12-lead ECG prior to catheter ablation and arrow bars for scale (in cm). (A) EWI isochrones of a patient with WPW and a left lateral AP. (B) EWI isochrones of a patient with PVCs originating from the left anterior papillary muscle. Views shown [top to bottom for (A) and (B)] include 3D-rendered ventricular isochrone in anterior view with single lead ECG obtained during EWI acquisition, coronal cross section of the 3D-rendered isochrone, and transverse cross section at the level of the valves. (C) EWI isochrones of a patient with a focal AT originating from the posterolateral high right atrium. (D) EWI isochrones of a patient with a mitral AFL. Views shown for (C) and (D) include anterior (left) and posterior (right) views of the 3D-rendered atrial isochrones with single lead ECG obtained during EWI acquisition displayed below.

Accessory pathways

Twelve patients with manifest pre-excited 12-lead ECGs were imaged using EWI prior to AP ablation. There were 3 left lateral, 1 left anterolateral, 4 posteroseptal, 1 right posterior, and 3 anteroseptal pathways. Catheter ablation was successful in 100% of cases, which was defined as loss of the delta wave on 12-lead ECG and inability to induce supraventricular tachycardia. Analysis of correct predictions using EWI was performed using the segmented template described in fig. S3. EWI correctly predicted 12/12 (100%) of the AP locations (Table 2A). Representative images of a left lateral AP including 2D- and 3D-rendered isochrones are shown in Fig. 3A and representative images of a posteroseptal AP including 2D- and 3D-rendered isochrones are shown in fig. S4.

Table 2. EWI and ECG results.

a.	*Overall Correct
	Diagnosis	EWI (N)	EWI % Correct	ECG (N)	ECG % Correct
	WPW	12	100	72	64
	PVC	11	91	60	80
	AT	7	100	36	50
	AFL	25	96	144	74
	AT and AFL combined	32	97	180	70
	Total	55	96	312	71
b.	**Direct Comparison of EWI vs. 12-Lead ECG
	Diagnosis	Odds Ratio (95% CI)			p
	WPW	9.1 (1.4-∞)			0.016
	PVC	2.25 (0.2-23.4)			0.497
	AT	7.53 (1.0-∞)			0.051
	AFL	8.2 (1.0-67.1)			0.049
	AT and AFL combined	13.1 (1.6-104.6)			0.015
	All Patients (unadjusted for diagnosis)	11.8 (2.2-63.2)			0.004
	All Patients (adjusted for diagnosis)	12.1 (2.3-63.2)			0.003
c.	Inter-observer Agreement of ECG reads between clinicians
	Diagnosis	Kappa (agreement)(52)
	WPW	0.33 (minimal)
	PVC	0.33 (minimal)
	AT	0.00 (none)
	AFL	0.37 (minimal)
	AT and AFL combined	0.36 (minimal)

^*EWI and ECG % correct include both exact segments and adjacent segments as described in the methods and fig. S3. For further characterization of EWI and ECG accuracy, please see Fig. 6.

^**Odds ratios for PVC, AFL, AT/AFL, and total patients performed with generalized linear mixed model. Odds ratios for WPW and AT calculated using exact logistic regression. Interpretation of Kappa values as per reference. ECGs (N) = # of cases with ECG available x 6 readers.

Premature ventricular complexes

Eleven patients were recruited prior to catheter ablation of PVCs and imaged with EWI. PVC locations confirmed after EPS included 4 septal right ventricular outflow tract (RVOT), 1 high posterior RVOT, 1 aorto-mitral continuity, 1 right coronary sinus of Valsalva, 1 epicardial LV summit, 1 left anterior papillary muscle, 1 posterior papillary muscle of the tricuspid valve, and 1 right ventricular septum. The mean PVC burden was 29 ± 2% (as determined by 24-hour Holter monitor or implanted cardiac monitor). Catheter ablation was successful, defined as absence and non-inducibility (absence with and without intravenous isoproterenol) of primary PVC morphology, in 10/11 (91%) cases. Analysis of correct predictions using EWI was performed using the segmented template (fig. S3), including single segments for the RVOT and left ventricular outflow tract (LVOT), respectively. EWI correctly identified 10/11 (91%) of PVC locations (Table 2A). No area of earliest activation could be determined on the isochrones for the single case for which EWI was unable to locate the PVC origin. In this case, the PVCs originated from the RVOT which was outside of the acquired EWI views, as transthoracic imaging of the RVOT is limited. Representative images from the patient with left anterior papillary muscle PVCs are shown in Fig. 3B, and 2D and 3D-rendered isochrones of both consecutive sinus and PVC beats prior to catheter ablation are shown in fig. S5 and movie S1.

Focal atrial tachycardia

Seven patients presented for focal AT ablation with the following locations identified on EPS: 2 high posterior right atrium, 1 right atrial septum, 1 crista terminalis, 1 low lateral right atrium, 1 left inferior pulmonary vein, and 1 left atrial roof tachycardia originating from an accessory pulmonary vein. Therefore 5 originated from the right atrium and 2 originated from the left atrium. The mean cycle length of the ATs was 362 ± 20 ms. Catheter ablation was successful in terminating 100% of AT cases. Analysis of correct predictions was performed as described in the methods. EWI correctly identified 7/7 (100%) of the AT locations (Table 2A). Representative images of the 3D-rendered isochrones of a patient with posterior right atrium AT are shown in Fig. 3C, and the corresponding four 2D isochrones of that same patient are included for more thorough visualization in fig. S6. A second patient with a right atrial free wall AT is shown in fig. S7, illustrating the difference between earliest activation in the lateral free wall (4-chamber view isochrone) vs. the posterior wall of the right atrium (3.5-chamber view isochrone from fig. S6).

Atrial flutter

Twenty-five patients presented for ablation of AFL. Twenty-one had typical cavotricuspid isthmus (CTI) dependent flutter and 4 had atypical atrial flutters originating from the left atrium. Of the atypical flutters, 2 were mitral AFLs, 1 was a left atrial roof AFL, and 1 was a left atrial anterior wall AFL. Mean cycle length of the tachycardias was 270 ± 6 ms. Catheter ablation was acutely successful in terminating 24/25 (96%) of AFL cases. EWI correctly identified the location of 24/25 (96%) of the AFL circuits (Table 2A). Representative images of 3D-rendered atrial isochrones in anterior and posterior views for a mitral AFL are shown in Fig. 3D. Multi-2D EWI atrial isochrone slices are also displayed for a CTI AFL (Fig. 4A) and the same mitral AFL (Fig. 4B).

Fig. 4. EWI isochrones of CTI and mitral flutter.

This figure shows the EWI isochrones of two patients presenting for ablation of AFL. 12-lead ECGs obtained prior to catheter ablation are shown. The isochrones are displayed as four multi-2D co-registered slices to illustrate the right and left atria from the anterior (left) and posterior (right) views. Red represents the earliest activated region and blue the latest. Arrow bars for scale (in cm) and single lead ECG obtained during EWI acquisition are shown below. See movies S2 and S3 for propagation videos. (A) EWI isochrones of a 62-year-old with a counterclockwise CTI AFL. The earliest activation is seen in the right atria around the tricuspid valve (red). (B) EWI isochrones of a 61-year-old with a mitral AFL. The area of activation is seen in the left atrium around the mitral annulus. See fig. S8 for other images of the multi-2D isochrone in a different orientation, further demonstrating the direction of propagation of the mitral AFL.

Clinical application of isochrone visualization

EWI isochrones can be co-registered to pre-ablation CT scans and 3D electroanatomic maps built with invasive catheter mapping, as shown in Fig. 5. The latter displays images of the 2D isochrones (Fig. 5A) and electroanatomic maps (Fig. 5B) of a left atrial roof AT, as well as the corresponding 12-lead ECG (Fig. 5C) and four-chamber isochrone co-registered to the pre-ablation CT scan (Fig. 5D). This patient with previous pulmonary vein ablation had an AT originating at the ostium of an accessory pulmonary vein in the LA roof. EWI isochrones can also be shown over a shorter time scale allowing for better characterization of the direction of propagation, as illustrated on the example of the mitral AFL in fig. S8. Finally, the isochrones can be played over time to further emphasize the activation propagation, as demonstrated by the videos of both CTI and mitral flutters (movie S2 and movie S3).

Fig. 5. EWI isochrones of a left atrial roof tachycardia.

This figure shows the EWI isochrones of a 64-year-old presenting for AT ablation after previous pulmonary vein isolation. On pre-ablation CT scan the patient was noted to have an accessory pulmonary vein originating from the left atrial roof. Red on the isochrones represents the earliest activated region and blue the latest. (A) Four 2D EWI isochrones of the atria illustrating earliest activation on the left atrial roof, arrow bars displaying the scale (in cm), and single lead ECG obtained during EWI acquisition. (B) CT scan of the left atrium (blue) and left ventricle (orange) alongside the electroanatomic map during the ablation procedure. On the electroanatomic map, the red arrow at the bottom illustrates the initial site of ablation, when the arrhythmia was believed to be originating from the mitral isthmus. The blue arrows at the top illustrate the successful site of ablation at the location of the accessory pulmonary vein. (C) 12-lead ECG obtained during EWI acquisition and prior to the EPS. (D) Four-chamber atrial isochrone overlaid onto the full cardiac CT scan. The blue arrow points to the accessory pulmonary vein and matches the location of the earliest activated region in the roof of the isochrone. Of note, transthoracic echocardiography was difficult in this patient, resulting in the 3.5- and 2-chamber views being more closely aligned than expected, and therefore preventing 3D rendering.

EWI compared to ECG analysis

Six board certified cardiac electrophysiologists were asked to predict the location of the arrhythmia by reading pre-operative 12-lead ECGs using any published algorithm they would have used in clinical practice. All electrophysiologists were blinded to EWI, 3D electroanatomical maps, and EPS reports. Three patients were excluded from ECG analysis due to lack of a 12-lead ECG in the appropriate rhythm (Fig. 1). ECG-based predictions were completed by all six electrophysiologists for the remaining 52 patients. Clinician interpretation of ECG correctly predicted 71% of the locations of arrhythmias, with minimal inter-observer agreement by diagnosis (Kappa values: WPW = 0.33; PVC = 0.33, AT = 0.00, AFL = 0.37). EWI was more accurate than 12-lead ECG for localization of arrhythmia or pathway origins in all patients (unadjusted for arrhythmia type: OR: 11.8; 95% CI: 2.2-63.2; P = 0.004; adjusted for arrythmia type: OR: 12.1; 95% CI: 2.3-63.2; P = 0.003). Accuracy by diagnosis (Table 2A), comparison of EWI with ECG (Table 2B), and inter-observer agreement of ECG reads (Table 2C) are all shown in Table 2. Heatmaps demonstrating predicted locations of the ventricular arrhythmias by EWI and cardiac electrophysiologists compared to localization with intracardiac mapping are shown in Fig. 6. There was no change in accuracy for EWI when predicted segments adjacent to the correct segments were disregarded. ECG accuracy for AP localization fell from 64% to 47% while PVC localization fell from 80% to 75% if only exact segments were considered correct.

Fig. 6. Heatmaps of EWI and ECG predictions vs. catheter-determined locations for AP and PVC localization.

This figure shows four heatmaps illustrating the EWI and ECG predictions for both (A) WPW and (B) PVC locations as compared to intracardiac mapping. Rows indicate EWI or ECG predictions, whereas the columns represent the intracardiac location of the pathway confirmed by invasive catheter electrophysiology mapping. The locations for both predictions and intracardiac mapping refer to the segmented cardiac map found in fig. S3. A “?” indicates that the reading physician or EWI was unable to make a prediction. The number in each cell indicates the number of predictions of the segment in the row as predicted by EWI or ECG respectively. Therefore, there is 1 prediction per case with EWI and 6 per case with ECG. Green indicates an exactly correct prediction. Yellow indicates a prediction in an adjacent cardiac segment as seen in fig. S3, which was counted correct in the results and as displayed in Table 2. Red indicates an incorrect prediction.

Discussion

This double-blinded study reports the clinical accuracy of multi-2D or 3D-rendered transthoracic EWI for noninvasive localization of arrhythmias in all four cardiac chambers. This was performed in an adult patient population with pre-existing cardiac disease including previous catheter ablations and/or other cardiovascular co-morbidities. In this study, the accuracy of EWI was higher than that of clinical diagnosis by electrophysiologists reading standard 12-lead ECGs. When used in conjunction with standard 12-lead ECG, EWI may be a valuable tool for diagnosis, clinical decision making, and treatment planning of patients with arrhythmias.

We previously published a study using EWI restricted to AP localization in minors with WPW, who were otherwise healthy with normal heart function and no other cardiac diseases (32). This study performed in an adult population shows a consistently high rate of EWI accuracy for localizing WPW, PVC, AT, and AFL in a more clinically diverse and heterogeneous patient population with known cardiovascular disease. On prior echocardiogram, 37.5% (n = 15) of patients had dilated left atria, 18% (n = 7) of patients had prior evidence of wall motion abnormality, 42.5% (n = 17) of patients had a decreased left ventricular ejection fraction, 20% (n = 8) of patients had a previously technically limited echocardiogram, and 13% (n = 7) of patients had previous catheter ablation procedures. Previous catheter ablation causing iatrogenic scar formation is known to make interpretation of the ECG more difficult (42). Pre-existing structural aberrations, such as scarring and wall motion abnormalities, did not diminish EWI localization accuracy. In this cohort, EWI was capable of successfully locating the site of interest in 96% (n = 53) of the 55 patients and was more accurate than 12-lead ECG. EWI was also capable of characterizing macro-reentrant circuits in addition to focal arrhythmias.

Standard 12-lead ECG has long been an important tool for localization of arrhythmias (2–9). However, the anatomic information provided by ECGs can be limited. Several methodologies have been proposed to increase the accuracy of the ECG, but these have had variable success. Furthermore, ECG for diagnosis and localization of arrhythmias have previously been shown to have a high degree of inter-observer variability, also seen in this study (10–16). The average accuracy of 12-lead ECG localization in this cohort was lower than previously published, as was the agreement between observers, which may have affected its comparison to EWI. This may be due to the usage of the standardized segmented map, which may have diminished the accuracy of localization, since some segments were close in proximity. However, given the high degree of accuracy of EWI, we believe that EWI used in conjunction with the 12-lead ECG may provide increased accuracy in the diagnosis of arrhythmias.

An advantage of EWI is the ease with which isochrones can clearly demarcate the earliest sites of interest along with direct anatomic visualization similar to standard transthoracic echocardiography. For example, 12-lead ECG may be limited in diagnosing arrhythmias from the posterior side of the heart, but EWI imaging can provide 3D-rendered anatomical information. EWI isochrones can also be imported and overlaid onto 3D electroanatomic mapping systems used in the electrophysiology laboratory. Although sites of intended treatment will necessarily be confirmed with intracardiac catheter mapping prior to ablation, an overlay of EWI isochrones on personalized patient anatomy in 3D electroanatomic mapping systems may potentially reduce procedural time by directing the treating electrophysiologist to the area of interest expeditiously. For example, prior knowledge of the location of origin of the focal AT at the juncture of the accessory pulmonary vein in the patient with left atrial roof tachycardia could have potentially prevented prolonged procedure and anesthesia times and decreased radiation exposure to the patient, who had previous pulmonary vein isolation. EWI could also be used immediately after unsuccessful ablation attempts, to help determine the reason for failure, by identifying changes to the arrhythmia after application of ablation lesions. Opportunities for integration of EWI in the clinical workflow are shown in fig. S9.

Another advantage of EWI is its use of pre-established infrastructure, using hardware that already exists in most echocardiography machines readily available in clinics and hospitals. EWI costs have a similar profile to standard 2D transthoracic echocardiography, and would be more cost effective than computed tomography or magnetic resonance imaging (33). Real time EWI could easily be integrated into existing standard clinical ultrasound imaging systems, since no additional hardware would be required. Real time implementation of EWI is being developed by our group. Its most recent version is able to process and generate the 3D-rendered isochrones in less than 10 minutes and is currently undergoing optimization and investigation. The real time prototype video as implemented on the research scanner computer with no additional offline processing required is shown on the example of a 26-year-old healthy male volunteer (movie S4).

Current techniques such as ECGI have been successful in noninvasively providing activation maps of arrhythmias at high spatial resolution and its use is growing in clinical settings. However, it can be cost- and time-inefficient, and expose the patient to ionizing radiation since CT is required for anatomical information. These drawbacks limit the use of ECGI as a test for the average patient presenting for evaluation of arrhythmia. In addition, unlike other imaging modalities which require in-depth user training and experience for interpretation, the isochrones generated by EWI provide information in a clear and easy-to-interpret manner for a general audience. This makes EWI a useful noninvasive tool that can be easily applied in everyday clinical practice and can be used to facilitate shared decision making between the operating electrophysiologist and patient prior to catheter ablation.

EWI relies on echocardiography and the presence of the rhythm of interest during acquisition. For this study, an ultrasound research scanner was used for functional imaging, not clinical structural anatomical imaging. Obtaining high quality data when imaging the atria for AFL and AT patients can be a challenge because the ultrasound wave attenuates the further it travels, such as towards the atria at the bottom of the field of view. In addition, echocardiography quality is operator dependent and has been previously shown to be limited by factors such as body mass index or pulmonary disease in up to 20% of patients (43). In our study, only 8%, 5 of the 60 imaged patients, had poor acoustic windows that limited echocardiography, indicating that EWI was successfully performed even in patients who are likely difficult to image due to comorbidities. Although univariate analysis of limited echocardiographic studies and predictors of EWI failure did not identify notable predictors, this study was neither powered nor designed for this analysis.

Secondly, even with high quality imaging, certain anatomical structures are difficult to thoroughly image with transthoracic echocardiography. This is most notable in this cohort with PVC localization in the right ventricular outflow tract (RVOT) and left ventricular outflow tract (LVOT). EWI can determine whether a PVC originates from the RVOT or LVOT, however providing more specific anatomical localization is difficult because the outflow tracts are not always thoroughly imaged with transthoracic echocardiography, particularly in the four apical views. Acquiring additional 2D EWI slices such as the parasternal long-axis or tricuspid tilt views, to specially image the RVOT region, could potentially provide more precise arrhythmia localization. More thorough characterization may also be better achieved through the use of true 3D volumetric ultrasound imaging, and more specifically EWI with trans-esophageal or intra-cardiac echocardiography. Nevertheless, transthoracic 2D EWI can direct the operator to a more specific area than 12-lead ECG alone.

This study was a prospective pilot analysis of clinical usage of the EWI technique, not a trial randomizing patients to EWI imaging before planned catheter ablation, and therefore the study did not test the effect of EWI on a specific clinical outcome such as reduced procedure, anesthesia, or fluoroscopy times. The study was performed at a single center and our patient population was selected. All patients in this study had been deemed suitable candidates for catheter ablation. Only the four selected cardiac arrhythmia diagnoses were included. We did not specifically determine endocardial or epicardial focality for the arrhythmias imaged. Although EWI was previously shown capable of distinguishing endocardial from epicardial origins in canine ventricles, none of the patients were found to have an epicardial focus in this cohort. Our results may not be applicable to all patients with similar arrhythmias. Given the sample size, we may have been underpowered to detect significant clinical factors associated with failure of EWI. Patients who consented to the study but had poor acoustic windows were excluded.

For future research, EWI with 3D ultrasound would allow visualization of the entire myocardial volume in a single heartbeat and could potentially increase the accuracy of EWI. Temporal co-registration of the four multi-2D isochrone views currently relies on manual p-wave or QRS origin selection on a single lead ECG, which corresponds to the earliest possible activation (0 ms). We ensured that the heart rhythm and ECG morphology were identical across all four views for each cycle prior to 3D-rendering. The temporal origin of the activation map was systematically selected on each view’s corresponding ECG by measuring an identical interval from the immediate R wave or p-wave peak. Repeating the origin selection process to have consistent isochrone starting times across the views is of utmost importance, especially for AFLs. Furthermore, in the cases of AFLs, we always selected the p-wave of interest on the ECG within the “sawtooth” p-wave pattern away from the QRS complex to avoid potential interference with ventricular signals. Future studies using high volume-rate 3D EWI would circumvent potential errors arising from the challenges of temporal multi-2D view co-registration, as well as out of plane motion, since whole heart cardiac electromechanical activity would be mapped in a single heartbeat. Our group has demonstrated that 3D EWI is feasible in open-chest canines in sinus rhythm, left ventricular pacing, and ventricular tachycardia (44, 45). Proof of concept of 3D EWI with a 32×32 matrix array in a clinical setting was established by our group on a healthy volunteer in sinus rhythm and in a cardiac resynchronization therapy patient during both right ventricular pacing only, and biventricular pacing (45). Finally, future studies will investigate (1) whether EWI can reduce procedure, anesthesia, and fluoroscopy times and/or improve outcomes, (2) reduce cost of ablation procedures for both focal and macroreentrant arrhythmias, and (3) utilize EWI during intracardiac echocardiography for more exact spatial resolution during invasive electrophysiology study.

In summary, EWI provides clear anatomic localization of the site of origin of arrhythmias that could be used for pre-procedure planning. Catheter ablations are a proven treatment method, but have inherent risk. For instance, transseptal puncture has risk of stroke and bleeding, ablation of accessory pathways near the AV node may cause heart block, and increased fluoroscopy times can carry complications from radiation exposure (46). EWI used as a clinical imaging modality could improve discussion with patients about potential treatment options and planning, while potentially reducing procedural risks and time.

Materials and Methods

Study design

This study was conducted to validate EWI’s ability to non-invasively localize cardiac arrhythmias in adult patients. The study was designed in a double blinded fashion to prevent bias in electroanatomic mapping and allow for EWI and ECG comparison. EWI was performed in patients who presented with WPW, PVC, AT and AFL for catheter ablation at the Columbia University Medical Center (CUMC) electrophysiology laboratory between July 22, 2016 and July 2, 2018. All patients presenting for ablation of the above arrhythmias were considered (see Fig. 1). Inclusion criteria included presence of the arrhythmia during EWI scans and ability to obtain required imaging views to perform EWI (adequate windows for ultrasonography). Patients were not excluded for any clinical comorbidities. Power analysis for EWI and ECG comparison assumed 80% ECG accuracy (based on published assumptions and mixed inclusion of arrhythmias) and 90% EWI accuracy with alpha of 0.05 and power of 80%, resulting in N=42. The CUMC Institutional Review Board approved this clinical study before the initiation of research activities and informed consent was obtained before each EWI scan. For the large animal model, the experimental protocol was in compliance with the Public Health Service Policy on Humane Care and Use of Laboratory Animals, and the LV-paced canine studies were conducted with approval of the Institutional Animal Care and Use Committee (IACUC) at Columbia University.

Electromechanical wave imaging

EWI relies on a high frame rate ultrasound flash sequence that emits a single diverging beam at 2000 frames/second (Fig. 2A) (24, 47). Two-dimensional ultrasound acquisitions in four apical echocardiographic views were performed at a 20-cm depth using a Vantage Research scanner (Verasonics Inc.) with a 64-element 2.5 MHz phased array (P4-2 ATL/Philips). The high frame rate radiofrequency (RF) data was acquired for 2 seconds, followed by a 1.5 second focused anatomical B-mode sequence at 30 frames/second. Lead II recordings of the ECG were obtained simultaneously by the system and synchronized with the EWI acquisitions. These ECG recordings were later used for temporal co-registration across EWI views by manually selecting the QRS and p-wave onset, for the ventricles and atria, respectively. The same heartbeat morphology was always selected on the separate multi-2D view ECGs to maintain consistent starting times across all four isochrones. The EWI RF signals were reconstructed with a standard delay and sum beamforming algorithm. Myocardial wall segmentation of the chambers of interest was performed manually on the first anatomical B-mode image (Fig. 2B) and automatically tracked in all subsequent frames of the cardiac cycle through systole (48). Motion was estimated axially with 1D cross-correlation tracking on the RF data with a window size of 10 wavelengths and 90% overlap, followed by a least-squares strain estimator with a 5 millimeter kernel to compute the electromechanical axial strain (Fig. 2C) (23, 49).

The quality of echocardiogram imaging is dependent on the skills of the operator. However, the EWI isochrone generation process does not depend on the patient geometry or on the skills of the technician holding the probe (19). The type of strain, whether axial or radial, is of no consequence as the only important factor is the sign change of the strain. For computational efficiency, activation times were manually selected on the incremental strain curves for a subset of about one hundred points, randomly and automatically chosen by down sampling the total number of pixels contained in the segmented mask. Activation times (t_act) were defined as the time-point of the first polarity change in inter-frame or incremental axial strain from relaxation to contraction (Fig. 2D), also known as zero-crossing (ZC), after the QRS and the p-wave onset, for the ventricles and atria respectively (19, 26). Representative examples of incremental axial strain curves are shown in fig. S10 with the corresponding ZC locations depending on the myocardial region and ECG interval of interest; for a healthy volunteer in normal sinus rhythm (fig. S10A) and for a typical CTI flutter (fig. S10B). During active contraction that follows isovolumic contraction, a change from lengthening to shortening in the axial direction of the myocardium is detected. Therefore, this corresponds to a positive-to-negative downward ZC (shortening) in the apical views, as the myocardium walls are mostly aligned with the ultrasound beam during contraction with the exception of the atrial roof. In that case, since the wall is orthogonal to the beam’s direction, the activation times correspond to the negative-to-positive upward ZC (thickening). A Delaunay triangulation-based cubic interpolation was then applied to the scattered activation time values to achieve a homogeneous isochrone pattern throughout the entire myocardium mask grid. The four resulting 2D isochrones or activation maps display the activation time (in milliseconds) from the point of interest (onset of p-wave or QRS) with earliest activation displayed in red and latest in blue (Fig. 2E).

After obtaining the multi-2D isochrones in the four apical standard views, the left ventricle median axis or longitudinal apex to base rotation axis of the probe (dotted black lines on Fig. 2E) was automatically detected on each view (50). Relative positions of the four 2D imaging planes were assumed to be organized as in the theoretical case with preset probe rotation angles: 60°clockwise between the 4-ch and 2-ch, 30° clockwise between the 4-ch and 3.5-ch, and finally 60°counter-clockwise between the 4-ch and 3-ch view. The four 2D isochrone slices were then automatically co-registered spatially around the LV longitudinal axis of rotation and a linear interpolation of the activation times was performed around the circumference (Fig. 2E and Fig. 2F).

Finally, 3D-rendering of the EWI maps were generated (Fig. 2F). The 3D-rendering algorithm runs in MATLAB, outputs the 3D-rendered isochrones as 3D arrays and the volumes are then imported into Amira for better visualization and manipulation (50). On top of static images, more dynamic visualization was achieved by playing the isochrones over time, enabling better characterization of macro-reentrant circuits (movie S2 and movie S3). The workflow of the entire EWI processing is shown in Fig. 2. Obtaining an EWI scan with the four multi-2D views took about 15 minutes in the pre-operative area on the day of the procedure, whereas the off-line processing of each EWI scan required about 90 minutes, including generation of both multi-2D and 3D-rendered isochrones (about 70 minutes for 2D isochrones only).

Double-blinded clinical study

All patients presenting for catheter ablation of one of the four included arrhythmias (WPW, PVC, AT and AFL) at the Columbia University Medical Center during study availability were considered. Patient characteristics were obtained from preoperative histories and medical records (Table 1 and table S1). Only during initial feasibility testing, the first thirteen patients were processed in a single blinded manner such that the EPS operators were blinded to EWI results, but EWI was not blinded to catheter ablation results. All data from the subsequent 42 patients were processed as a double-blinded study. The trained sonographer and the engineers who processed the EWI were blinded to the 12-lead ECG prior to each patient’s catheter ablation, the EPS results and the 3D electroanatomical maps. The operating electrophysiologists were blinded to all EWI results. Six board certified electrophysiologists not involved in the patient’s care were blinded from EWI and 3D electroanatomic maps and asked to use each patient’s 12-lead ECG prior to ablation to predict the arrhythmia site of origin or location of the AP. They were allowed to use any preferred algorithm for localization.

All patients with an AP had manifest ventricular pre-excitation on their resting ECG. Patients presenting for PVC ablation had monomorphic PVCs, and those presenting for AFL or AT ablation presented in those respective rhythms during EWI. For AP and PVC locations, EWI and ECG readers used a standardized segmented map of the ventricles with 21 anatomic locations to predict the origin (fig. S3). This map was designed for this study prior to patient enrollment, and is similar to the 17-segment model developed by the American Heart Association recently applied to ventricular tachycardia localization, with the addition of right ventricular segments (51). Reads were considered correct if predictions fell in the exact segment or in a directly adjacent segment to the actual location of the arrhythmia. For patients with AT or AFL, both clinicians and EWI assessed whether the arrhythmia was a typical CTI AFL, or of other right atrial versus left atrial origin (fig. S3). EWI localizations for all diagnoses were determined based on the earliest activated regions on the 2D isochrones prior to 3D rendering. Results of EWI and ECG analysis were compared directly to the 3D electroanatomical maps and the site of successful ablation.

EPS and ablation

Clinical electrophysiology studies were performed using standard equipment and electroanatomic mapping (CARTO, Biosense Webster or EnSite, Abbott Medical Inc.). After obtaining vascular access, multi-electrode catheters were positioned under direct fluoroscopy. A surface ECG was recorded prior to ablation. The average procedure time in this cohort was 147±60 minutes with an average fluoroscopy time of 20.0 ± 14.1 minutes. Entrainment maneuvers, activation sequence mapping and 3D electroanatomical mapping were performed for patients presenting with AFL, AT, or a conducting AP participating in a clinical arrhythmia. For patients presenting for PVC or WPW ablation with manifest ventricular pre-excitation, the ablation site was determined by the earliest activation during electroanatomic mapping either in sinus or ventricularly paced rhythm. Electrophysiology study and ablation reports for the 55 included patients and 12-lead ECGs available for 52 patients are presented in the Supplemental Material.

Canine study

To investigate the effect of multi-2D sampling on the spatial resolution of 3D-rendered EWI isochrones, two open-chest experiments were performed on LV-paced canines. The two male mongrel dogs (age = 9 and 7 month-old, weight = 29.6 and 28 kg) were anesthetized with an intravenous injection of propofol (4.4 mg/kg) and sustained under a mixture of inhaled oxygen and isoflurane by a rate- and volume-regulated mechanical ventilator (1%𢀓5%). A lateral thoracotomy procedure was used to expose the heart; two ribs were removed and the pericardium was incised for placement of the pacing electrodes. The bipolar electrodes were sutured externally onto the LV epicardial surface of the canine hearts and sent the following pacing signal: 1 V amplitude, 5 ms pulse width and 500 ms cycle length).For the first experiment the canine’s LV was paced at 5 different locations and the usual four EWI apical views were acquired each time by the sonographer (fig. S1). For the second dog the LV was paced at a single location but additional apical multi-2D views were acquired (fig. S2). A robotic arm was used in the latter to accurately measure the rotation angle of the probe between the additional EWI planes: either 6 evenly spaced slices by 30-degree angles or 12 evenly spaced slices by 15-degree angles (fig. S2A). These two animals were used for different cardiac studies, but procedure durations for this particular multi-2D EWI sampling study (excluding thoracotomy) spanned over 1.5 hours and 2.5 hours for dogs 1 and 2 respectively. Finally once all procedures were completed, the canines were euthanized by a lethal IV injection of Euthasol (5mL) while still under deep isoflurane anesthesia (5%).

Statistical analysis

Data were expressed as frequency (%) or mean ± standard error of the mean as appropriate. Variability analysis for ECG interpretations was performed using Light’s Kappa method (52). Odds ratios for comparison of EWI to ECG were achieved using a generalized linear mixed model, except when prevented by separation, in which case exact logistic regression was used. Univariate logistic regression, chi square tests, or Fisher’s exact tests were used for univariate analysis as appropriate. Variables reaching P <0.10 in univariate analysis were included in multivariate analysis. Multivariate analysis was performed with logistic regression. Statistical analysis was performed using both SPSS statistical software (Version 24, IBM corp.) and STATA Statistics / Data Analysis (Version 15, Stata Corp.).

Supplementary Material

Supplemental Material EP

Electrophysiology study and ablation reports for the 55 included patients 12-lead ECGs available for 52 patients

Supplemental Material

Fig. S1. Spatial resolution of standard 3D-rendered EWI in an open-chest canine with five different LV pacing locations

Fig. S2. Multi-2D sampling effect on the resolution of 3D-rendered EWI in an open-chest canine with a single LV pacing location

Fig. S3. Standardized segmented map of the heart

Fig. S4. EWI isochrones of a posteroseptal AP

Fig. S5. EWI isochrones of a sinus beat and its consecutive PVC beat prior to catheter ablation

Fig. S6. EWI isochrones of a right posterior atrial tachycardia

Fig. S7. EWI isochrones of a right lateral free wall atrial tachycardia

Fig. S8. EWI isochrones of a mitral flutter displaying direction of propagation

Fig. S9. Opportunities for EWI integration into clinical workflow

Fig. S10. Representative examples of incremental axial strain curves with the corresponding zero-crossing (ZC) locations depending on the myocardial region and ECG interval of interest

Table S1. Patient characteristics by subgroup

Table S2. Predictors of limited echocardiography windows and EWI failure

Supplemental Video 1

Movie S1. Video of PVC beat and sinus beat activation propagation

Supplemental Video 2

Movie S2. Video of CTI flutter activation propagation

Supplemental Video 3

Movie S3. Video of mitral flutter activation propagation

Supplemental Video 4

Movie S4. Video of real time EWI prototype