Three-Dimensional Noninvasive Imaging of Ventricular Arrhythmias in Patients with Premature Ventricular Contractions

Abstract

Objective

Noninvasive imaging of cardiac electrical activity promises to provide important information regarding the underlying arrhythmic substrates for successful ablation intervention and further understanding of the mechanism of such lethal disease. The aim of this study is to evaluate the performance of a novel three-dimensional (3D) cardiac activation imaging technique to noninvasively localize and image origins of focal ventricular arrhythmias in patients undergoing radio frequency ablation.

Methods

Pre-procedural ECG gated contrast enhanced cardiac CT images and body surface potential maps were collected from 13 patients within a week prior to the ablation. The electrical activation images were estimated over the 3D myocardium using a cardiac electric sparse imaging technique, and compared with CARTO activation maps and the ablation sites in the same patients.

Results

Noninvasively-imaged activation sequences were consistent with the CARTO mapping results with an average correlation coefficient of 0.79, average relative error of 0.19 and average relative resolution error of 0.017. The imaged initiation sites of premature ventricular contractions (PVCs) were, on average, within 8mm of the last successful ablation site and within 3mm of the nearest ablation site.

Conclusion

The present results demonstrate the excellent performance of the 3D cardiac activation imaging technique in imaging the activation sequence associated with PVC, and localizing the initial sites of focal ventricular arrhythmias in patients. These promising results suggest that the 3D cardiac activation imaging technique may become a useful tool for aiding clinical diagnosis and management of ventricular arrhythmias.

I. INTRODUCTION

Approximately 400,000 sudden cardiac deaths are reported annually in the United States alone, according to the American Heart Association [1]. In addition to the medication routinely administered to suppress such syndromes, radio frequency (RF) ablation has been increasingly performed to manage ventricular arrhythmia [2]. In recent years, effort and success has been made on identifying ventricular arrhythmia along the timeline of ECG recordings [3], [4]. However, in clinical practice, the success rate of spatially localizing PVC is still as low as 60% [5]. While catheter-based endocardial mapping approaches have shown the promise of offering a minimally invasive means of localizing and mapping cardiac electrical activity, these catheter intervention approaches are time-consuming, limited in hemodynamically unstable arrhythmias, and the success rate in patients with structural heart disease can still be as low as 60% [6]. The knowledge on location and propagation patterns of arrhythmia is crucial for a successful ablation. However, the invasive intra-cardiac mapping technology demands prolonged periods in the electrophysiology (EP) lab and hospitalization.

Efforts have been made to reconstruct cardiac electrical activity from the body surface potential maps (BSPMs). Various approaches have been proposed previously in different scopes, such as moving dipole solutions [7], [8], epicardial potential imaging [9]–[14] and heart surface activation imaging [14]–[21]. Evaluation studies on the approaches involve animal experiments and human studies, displaying potential for assisting in clinical practice. However, the origin of the arrhythmic activity may not always be localized on the endocardial or epicardial surfaces but can also be found in transmural myocardium [5], [23]. The detailed knowledge of intramural electrical activity has proven to be important to the planning of clinical management in catheter ablation [5]. There is a clinical need for a non-invasive imaging approach that can estimate and visualize the transmural electrical activity.

Further endeavors have been made to image the electrical activities throughout the three-dimensional (3D) ventricles [11], [24]–[33]. Results from previous studies have demonstrated that the 3D cardiac electrical imaging technique is capable of localizing the 3D ectopic initiation site of different types of arrhythmia [25], [34], [35]. Yet clinical challenges remain that demand non-invasive imaging techniques capable of achieving high spatiotemporal resolution while avoiding limitations from physiological presumptions. A very recent development in the field of noninvasive 3D cardiac electrical imaging – the Cardiac Electrical Sparse Imaging (CESI) method - has been proposed to meet these demands and has demonstrated improved performance in computer simulations and pilot pacing studies in two small animals [36].

In the present study, we aim to test the hypothesis that the CESI technique can localize and image initiation sites of focal ventricular arrhythmias in patients in a clinical environment. We report the first clinical study, to our knowledge, for imaging and localizing the origin of ventricular arrhythmia using 3D cardiac electrical activation imaging in patients undergoing ablative therapy, and compare these results with the intracardiac mapping data and ablation outcome.

II. METHODS

A. Three-dimensional Cardiac Electrical Imaging

In the process of depolarization, the transmembrane potential (TMP) rapidly rises from the −90 mV resting state to the 0 mV depolarized plateau potential level. Despite the individual variations in the resting and excited potentials, the transmembrane current flow, induced by the rapid change in TMPs, synchronizes with the cell activation while remaining silent during the remainder of the cardiac cycle. This phenomenon indicates strong temporal sparse properties that can be incorporated in the reconstruction of cardiac electrical activities.

In this study, we used a recently developed CESI technique. The mathematical formulation and detailed methodology have been described previously [36]. Based on the bidomain theory [37], at location r of any time instant t, the mathematical relationship between extracellular potentials and the intra-cardiac current density can be described as:

$$\nabla \cdot \left[\left(G_i\left(r\right)+G_e\left(r\right)\right)\nabla {\mathrm{\Phi }}_e\left(r,t\right)\right]=\nabla \cdot J_{eq}\left(r,t\right)$$ (1)

where G_e (r) and G_i(r) are the extracellular and intracellular effective conductivity tensors and ϕ is the extracellular potential at location r and time instant t.

With linearization techniques of modern numerical methods such as the Boundary Element Method (BEM) [38], the above governing equation can be reformatted as:

$$\overrightarrow{\mathrm{\Phi }}=L\overrightarrow{J}$$ (2)

where L stands for the transfer matrix and ϕ represents body surface potentials and equivalent current density, respectively.

Directly solving for the current density from observations on the body surface has been challenging due to the limited number of observations compared with the massive amount of information inside the myocardial tissue without prior knowledge. However, by utilizing the temporal sparse concept, a universal property on the behavior of cardiac electrophysiology, intra-cardiac electricity can be reconstructed with high spatiotemporal resolution by solving:

$$\begin{array}{l}{\widehat{J}}_T=\arg min\left({\Vert {\overrightarrow{J}}_T-L_T{\overrightarrow{\mathrm{\Phi }}}_T\Vert }_2^2\right)\\ s.t.{\displaystyle \sum _t^T{W}_{t,i}}{\Vert {\overrightarrow{J}}_{t,i}\Vert }_2^1<\mu E_i\mathit{for}\mathit{all}i\end{array}$$ (3)

where W represents the soft temporal weights of time instant t at location i. E represents the estimated energy of equivalent current density within the time window T at location i. This is a strong convex problem, which can be solved by applying convex optimization techniques. The problem is comprised of two parts: the norm residual term (top) and the constraints (bottom). While the residual term demands concurrence between the imaged electrical activity and body surface observations, the constraints mathematically describe the dynamic properties of the imagined results. By incorporating temporal sparse terms as shown in the formulation of the constraint term, the uniqueness that plagues the inverse problem is overcome and a solution can be generated to pinpoint the time of activation of each myocardial voxel.

Figure 1 illustrates the general principle and procedures of cardiac electrical activation imaging. The patient’s individualized volume conductor can be built numerically based on the torso geometry obtained by means of structural tomography imaging such as computed tomography (CT) or magnetic resonance imaging (MRI). By coupling the functional information from recorded BSPM and the geometrical information from CT or MRI, cardiac electrical imaging methods can reconstruct the cardiac electrical activity, revealing vital information regarding the arrhythmias such as initiation sites and propagation patterns. The temporal sparse promoting technique incorporates the cardiac electrophysiology into the reconstruction and can image the cardiac electrical activation sequence with improved accuracy, robustness and temporal resolution compared to the conventional minimum-norm based approaches

Fig 1.

Illustrations of cardiac electrical imaging. When a cell activates, the transmembrane potential changes rapidly and generates electrical activity that can travel through the body and be detected by ECG electrodes. With numerical methods such as the boundary element method (BEM), the electrical activity can be reconstructed in the form of equivalent current densities inside the myocardium to visualize the activation pattern of the cardiac depolarization process.

B. Study Protocol

A total of 13 patients with ventricular arrhythmias undergoing EP study with catheter ablation treatment participated in this study and all provided written consented. All experimental protocols were approved by the Institutional Review Boards of the University of Minnesota and of the Shanghai Ruijin Hospital. All patients were diagnosed with idiopathic premature ventricular complex (PVC) or ventricular tachycardia (VT). Twenty-four hour Holter recording was deployed to all patients after being admitted to the hospital. Preliminary diagnosis indicated that the patients suffer from frequent PVC (around or above 10,000 PVC/day). Detailed information of the patients is summarized in Table 1. The schematic of the study is shown in Figure 2. Body surface mapping was performed within one week of the EP study with ablation. Up to 208 electrodes arranged in strips (ActiveTwo system, BioSemi V.O.F., Amsterdam, the Netherlands.) were evenly distributed to cover the patient’s chest and back. Locations of the electrodes and the anatomical landmarks were digitized using an electromagnetic digitizer (Fastrak, Polhemus Inc, Colchester, VT, USA). ECGs were recorded at a sampling rate of 2 kHz using a 24 bit analog to digital converter. Thoracic (0.78×0.78×5 mm) and ECG gated (70% R-R interval) contrast-enhanced cardiac (0.39×0.39×0.75 mm) axial CT imaging were performed on all patients preceding the ablative procedure within a week of the mapping study.

TABLE I.

SUMMARIZED STATISTICS ON PATIENTS AND MODELING DETAILS.

Patient No	Gender	Age	PVC/24h	Myocardium Grids	Body Surface Grids	PVC origin
1	M	43	10000	13248	3566	RVOT
2	M	53	25630	12114	3478	RVOT
3	F	54	13411	11663	3145	RVOT
4	M	37	27161	13011	3235	RVOT
5	F	41	15160	11840	3387	RVOT
6	F	71	21100	12589	3233	RV
7	F	42	26337	12773	3171	RVOT
8	M	62	20000	12011	3412	LV
9	M	43	39476	13222	3517	RVOT
10	F	47	10000	12544	3227	RVOT
11	F	33	51226	10212	3111	RVOT
12	F	61	8470	11255	3199	RVOT
13	F	55	31062	12014	3308	RVOT

Fig 2.

Schematic diagram of the study. A: patients; B: body surface ECG collected pre-surgically including the arrhythmias; C: CT scan with torso and cardiac geometry; D: EP study with catheter ablation by intra-cardiac mapping technique; E: BSPM of the ectopic beats isolated from the ECG recording. F: boundary element model constructed according to the CT geometry; G: 3D co-registration of CARTO geometry and CT endocardium; H: CESI activation in 3D; I: quantitative comparisons between the CARTO mapped activation pattern and the CESI activation pattern on corresponding regions.

The EP study with ablation was conducted using a contact endocardial mapping system (CARTO3 system, Biosense Webster, Diamond Bar, CA, USA). Local activation time (LAT) from CARTO on the endocardium was measured sequentially across the entire or partial endocardium surrounding the suspected foci. Ablation was performed on the suspected foci of the arrhythmic activity until all ectopies were eliminated. All patients were symptom-free after the procedure and were discharged within one week.

C. Data Analysis

Segmentation of patients’ CT images and surface triangulations were carried out by using CURRY 6.0 software (Compumedics, Charlotte, NC). Triangular meshes of the realistic geometry (torso, lung, epicardium and myocardium) and ECG recordings were exported to the customized CESI software developed in Matlab 2010a (Math Works, Natick, MA). Detailed modeling statistics are summarized in Table 1. Activation sequences were imaged using the 3D inverse reconstruction algorithm to visualize the electrical activity during the selected arrhythmic beat of interest.

LAT maps and the intra-cardiac geometry were extracted from the EP study and co-registered to the corresponding CT images using the digitized anatomical landmarks. The intra-cardiac geometry collected from CARTO was co-registered to the corresponding endocardium using a four degree-of-freedom (locations and scale) rigid body registration method with fixed orientation [39] where the CARTO endocardium geometries are automatically aligned to the segmented CT geometry based on similarities in shape and locations without rotation. The imaged activation on the endocardial surface was also extracted from the 3D myocardium and aligned with geometry constructed by CARTO.

Six quantitative indices - Localization Error (LE), Nearest Ablation Localization Error (NALE), Correlation Coefficient (CC), Relative Error (RE) and Relative Resolution Error (RRE) - were used in the data analysis to evaluate the accuracy and clinical potential of the imaging technique. LE is defined as the distance between the imaged activation and the last successful ablation site. NALE is defined as the distance between the imaged initiation site and the nearest ablation site. CC and RE are utilized to assess the concordance and discrepancy between the invasively mapped and non-invasively imaged activation patterns on the endocardium, defined as:

$$CC=\frac{{\displaystyle \sum _i\left(A{T}_i-\overline{AT}\right)•\left(M{T}_i-\overline{MT}\right)}}{\sqrt{{\displaystyle \sum _i{\left(A{T}_i-\overline{AT}\right)}^2}•\sqrt{{\displaystyle \sum _i{\left(M{T}_i-\overline{MT}\right)}^2}}}}$$ (4)

$$RE=\sqrt{\frac{{\displaystyle \sum _i{\left(A{T}_i-M{T}_i\right)}^2}}{{\displaystyle \sum _{i}M{T}^2}}}$$ (5)

where AT_i and MT_i represent the imaged and measured activation times at the i th grid point while $\overline{\text{AT}}$ and $\overline{\text{MT}}$ represent their average on all grid points. RRE is employed to measure the loss temporal resolution, and is defined as:

$$\mathit{RRE}=\frac{\left|T_M-T_I\right|}{T_{\mathit{total}}}$$ (7)

where T_M and T_i represent the total activation time of measured and imaged activation sequences, respectively, and T_Total represents the total activation time from the selected ECG segment.

In addition to the comparison of activation sequences globally, the initial phase of the ectopic activation was also made to the corresponding CC and RE at the earliest 10 ms of the activation. The initiation sites between the imaged and the measured activation sequences are on average 7.9mm. Therefore the first 10 ms of the measured activation and the first 10 ms of the imaged activation are only partially overlapped, in which the calculation of the CC and RE is conducted.

Additional investigation was pursued regarding how the 3D imaging technique compared with the conventional 12-lead ECG approach in identifying the location of the ectopic initiations. 12-lead ECG electrode locations were decided based on CT geometry and the signal waveforms were interpolated from the BSPM. Single moving dipole localization was performed on all patient data at the initiation of the PVC beats. The comparison of these two paradigms is shown in figure 3. Infinite homogenous models are preferred to BEM based models in the dipole localization process for a typical 12 lead ECG interpretation process, commonly because sources of structural information such as CT or Ultrasound are absent.

Fig 3.

The schematic diagram and the statistics of quantitative comparisons between 12-Lead dipole localization and 3D imaging. Panel A: schematic diagram of comparing 3D imaging localizations and 12 Lead based dipole localizations. Panel B: summarized statistics of the 3D imaging localization and 12 lead ECG dipole localization.

III. RESULTS

A. Patient Population and Summarized Statistics

Ten PVC beats were randomly selected from each of the 13 patients for non-invasive imaging. Table 1 provides a summary of patient basic information, modeling statistics and diagnosed region of PVC initiation. The CC, RE, LE, NALE and RRE for each patient were computed and are summarized in Table 2. An average CC of 0.79 and RE of 0.19 indicate a strong concordance between the imaged activation sequence and the intra-cardiac measured LAT map. Furthermore, the concordance between the imaged and the measured activation in the early phase of the activation is also quantified in CC of 0.67 and RE of 0.33. The LE lies below 8 mm and the NALE below 3mm on average, indicating the imaged localization is comparable with the intra-cardiac mapping technique. RRE, as shown in Table 2, is well controlled below 3% and the temporal resolution loss is minimal. CC, RE and RRE of patient 7 are excluded from the summarized statistics with only LE and NALE utilized due to the inconsistency between the invasive mapping results and the ablation outcomes. The variation in the statistics may be due to the differences in initiation locations and consequentially the quality of the signal and complexity of the corresponding structure. The averaged point-to-point registration error is 2.2mm for all patients.

TABLE II.

STATISTICS SUMMARIZED BETWEEN CARTO MAPPED AND IMAGED ACTIVATION SEQUENCE ON THE CARTO MAPPED ENDOCARDIUM.

Patient	CC	RE	LE	NALE	RRE
1	0.78	0.16	8.1	2.7	0.01
2	0.8	0.17	7.5	4.5	0.02
3	0.78	0.21	7.3	1	0.03
4	0.81	0.2	9.2	1.3	0.01
5	0.82	0.18	7.7	4	0
6	0.81	0.16	8.6	4.2	0.03
7	N/A	N/A	6.2	1.7	N/A
8	0.75	0.23	6.6	4.1	0.02
9	0.8	0.18	7.1	4.5	0.01
10	0.81	0.16	8.2	2	0
11	0.77	0.19	9	1.9	0.01
12	0.77	0.22	9.2	3.5	0.02
13	0.78	0.23	8.5	1.5	0.01
Mean±sd	0.79±0.02	0.19±0.03	7.9±0.9	2.8±1.3	0.014±0.01

B. Focal Arrhythmias in RV and LV

Figure 4 displays examples of CESI activation imaging on ectopic beats initiated from the right and left ventricles from patient 6 and patient 8. Panel A shows an example of a PVC beat originating in the RV near the RVOT. The focal pattern can be observed from the 3D activation sequence that initiated in RV and dissolved at the LV wall. As the activation propagates from the RV free wall, a delay occurrs when it enters the LV, in contrast to a relatively faster propagation observed in RV. On the endocardium shown on the right, the focal pattern is well captured in the imaged results and is consistent with the CARTO activation maps. The intra-cardiac mapped activation sequence displays a relatively diffused pattern in the early activated area, in contrast to a more focused initiation pattern imaged. In panel B, the focal pattern originates in the left anterior wall and is also well imaged globally from the LV to the RV free wall. The initiation lies in the LV close to the apical anterior region. In the endocardium view, the localized tissues identified by the invasive and non-invasive techniques agree well with each other. By comparing the two, one can see that the imaged activation sequence has a clear focal activation pattern whereas the measured one holds more irregularity.

Fig 4.

Examples of imaging activation sequences during ectopic beats originating from the RV (Panel A) and the LV (Panel B) in two patients. Black circles represent the last successful ablation site. The white star indicates the site of initiation localized by CESI. The left column shows the 3D volumetric view of the myocardium activation. The cross-sections along the axial direction are presented in the middle column. In the right column, the endocardial surface is extracted from the 3D imaged activations and compared with those measured by CARTO.

C. Focal Arrhythmias in RVOT

Figure 5 shows two more examples of the imaging results of PVCs from patient 3 and patient 9. In Panel A, the PVC beat initiated near the RV free wall while the beat in panel B initiated at the septal side. In Panel A, activation initiated from the RVOT close to the RV free wall, traveled through the RV and terminated at the LV free wall. Initiating close to the septum, the activation traveled faster on the RV free wall than it did across the septum. In Panel B, different from the previous case, the PVC beat originated at the RVOT close to the septum, traveled through the septum and the two ventricles, and dissolved at RV and LV respectively. By comparing the limb-lead ECG of the two examples, it can be observed that the ECG pattern of the majority of the beat is similar in both cases but differences exist in the termination of the beat. In panel A, the activation terminated in the basal free wall, generating a propagation direction different from the bulk of the beat. Therefore, a small turning point can be found at the end of the limb lead QRS waveform. In contrast, the activation sequence in panel B, although terminating in two locations, has a generally uniform propagation pattern and reflects the smoother waveform found in the limb lead ECG.

Fig 5.

Examples of PVC ectopic beats originating near the RV free wall (Panel A) and septum (Panel B) in two patients. Black circle represents the last ablated site recorded. The white star indicates the CESI imaged initiation. The left column shows the 3D volumetric view of the myocardium activation. The cross-sections along the axial direction are presented in the middle column. In the right column, the endocardial surface is extracted from the 3D imaged activations and compared with those measured by CARTO.

A comparison of localization between 3D activation imaging and conventional 12-lead ECG dipole localization was also conducted, the results of which are summarized in figure 3. The 3D imaging method demonstrates a clearly superior capability in pinpointing the initiation of the ectopies.

IV. DISCUSSION

This study represents, to our knowledge, the first clinical investigation to demonstrate the performance and the applicability of the 3D cardiac activation imaging approach in patients who underwent EP study with ablation. This is also the first time, to our knowledge, that 3D activation imaging techniques have shown promise to map initiation sites of ventricular arrhythmias that are comparable to invasive CARTO results in a clinical setting. As a substantial extension from the previous numerical and pilot animal studies, the 3D activation imaging technique has been further fine-tuned and applied in a realistic clinical environment. The results of this clinical study presented here demonstrate the potential of this technique to reveal critical knowledge about the arrhythmic activity and directly assist in catheter ablation. In general, the focal pattern of the PVC activation propagation is accurately captured non-invasively and the imaged initiation site also demonstrates high concordance to the ablation sites as well as the procedural outcomes. On average, a CC of 0.79 and an RE of 0.19 were obtained in 12 out of 13 patients, indicating a high accuracy of the noninvasive imaging technique to directly estimate and visualize the arrhythmic activation pattern inside the heart. The corresponding CC and RE at the early phase of the activation are 0.67 and 0.35 respectively. As we can see, the statistics slightly deteriorated compared to those values when a larger endocardium was mapped. Note that the variations in the statistics are due to a smaller area and consequentially increased influence from CARTO-CT registration error. The imaged initiation sites were localized on average within 8 mm of the last successful ablation site and within less than 3mm to the nearest ablation site. The temporal resolution is well preserved with an RRE of less than 2 percent in general. The excellent performance of 3D imaging techniques revealed in the evaluation demonstrates that CESI is potentially able to provide a close match to invasive mapping techniques not only on endocardium but further throughout the three-dimensional myocardium on a beat-to-beat basis.

Imaging ventricular arrhythmic activation in a 3D volume poses a major challenge and, to our knowledge, the CESI method represents the best performance in imaging 3D cardiac activation according to the literature. CESI has been tested in simulations and pilot animal studies in 2 small animals using an artificial pacing protocol.. There has been no literature reporting clinical studies using the CESI method and the present study provides important data to establish the capability of the CESI methodology for imaging 3D ventricular arrhythmia as applied to patients. As a medical technology, the performance – accuracy and reliability – in a clinical environment of an imaging method is extremely important as its true battlefield. The complicated electrical conditions and unknown patient physiologies can, and have been shown to severely compromise imaging capability in contrast to the controlled simulation environment. Therefore, the clinical study has importance as a key milestone to wider clinical applications. Furthermore, to our knowledge, this is the first report to validate 3D ventricular activation imaging in patients with arrhythmia, aside from using the CESI method. Our results not only demonstrated the validity of the CESI method, but also the merits of 3D ventricular activation imaging in patients with arrhythmias, which, in our opinion, represents a novel original contribution to advance the state of the art of the field.

It is important to rigorously validate a non-invasive cardiac imaging technique using established clinical procedures such as CARTO mapping and ablation outcome. The present study uses quantitative evaluation of a noninvasive activation imaging technique regarding both the concordance to the ablation outcome and the agreement with intra-cardiac mapping activation sequences. The results indicate that the imaged initiation lies on average within 8 mm from the last successful ablation site. This would enable electrophysiologists to perform a quick and economical initial localization of the PVC foci before a full-fledged ablation procedure is conducted. Despite the 8 mm distance between the imaged initiation sites and the last successful ablation site, the imaged sites are on average 2.7 mm away from the nearest ablation site. While the exact role of each ablation site to terminate arrhythmia is debatable, the quantitative assessment we have provided in this study in terms of distance to the most recent ablation site and nearest ablation site, shall provide useful data for future investigations and also represent reasonable metrics for assessing the localization performance. With the aid of the 3D activation imaging technique, the area to be mapped with invasive techniques and the ablation area can be further reduced. As a medical imaging technology, the capability to properly function in a clinical environment is of great importance in its evaluation.

Unlike the studies in a controlled laboratory, the ground truth in a clinical setup is not always accessible as the major goal of the practice is to properly treat the patient rather than understand the physiology or pathology. Admittedly, intra-cardiac electroanatomical mapping is less ideal in terms of rigorousness than the 3D transmural mapping as we have been using over the past decade. The CARTO activation maps are made based on direct contact measurement of cardiac electrical signal on endocardium, making it the best available clinical approach for comparison with the noninvasive imaging method.

In our efforts to demonstrate the clinical performance of the imaging method, we have presented different aspects of evaluation to compare the imaging results with the most relevant clinical evidence accessible – the ablation sites. While there is more than one ablation site and it is virtually impossible to identify exactly which one is responsible for terminating the targeted arrhythmia, strong concordance has been demonstrated in the paper between the 10-min ECG imaging technique and hours-long ablation procedure. This strong co-localization displays the potential to further optimize the clinical workflow in the management of ventricular arrhythmias.

Taking extra ablation around the suspected ablation site to ensure the elimination of the targeted malfunctioning tissue or conduction pathway is far from uncommon in clinical practice. As discussed above, it is difficult to verify which single or combination of ablation attempts terminated the arrhythmia. This is the very reason we used both last ablation site and the nearest ablation site to provide different perspectives to help the readers understand the clinical relevance and performance of the imaging technique. In this paper, we report the data on both metrics to assess the performance of our imaging method, instead of suggesting that either the last or nearest ablation site should be adopted as the metric assessing performance of ablation.

Efforts have been made to visualize the cardiac electrical activities based on the body surface recordings [9], [11], [13], [15], [19], [29], [31]–[34], [36], [40]–[49]. CESI employs a physical model in cardiac imaging and the construction of a distributed current density source model. This enables the technique to image electrical activities throughout the 3D myocardium without relying on any presumed physiological knowledge, and therefore is intrinsically robust in various pathological conditions. As a result, CESI has the potential to directly identify the activation pattern and the regions of interest, such as ectopic foci or reentrant pathways, regardless of their positions or depths inside the heart. In Figure 4, both of the PVC beats are initiated in the RVOT but on the opposite sides of the outflow track: one near the RV free wall and the other near the septum. The BSPM and the limb lead ECG patterns are similar and the CARTO also identified a similar focal pattern with only slight differences in initiation locations. However, the 3D cardiac activation imaging technique was able to discriminate not only the initiation sites, but also different propagation patterns closely related to the detailed structure of the heart. The ability to identify the different propagation pathways based on similar ECG recordings demonstrates the capability of CESI to effectively integrate functional and structural information to image the arrhythmic mechanisms which play a major role in the treatment of more complicated arrhythmias.

The reconstruction algorithm in the imaging method uses a physical model as the energy observed on the body surface is inversely reconstructed into electrical activities inside the myocardium. Therefore, at the very beginning and very end phase of the activation in PVC, unlike the sinus rhythm, the observed energy is relatively low and therefore slightly shortens the QRS duration. The QRS duration is slightly shorter than the usual 12 Lead ECG measurements but the impact on the overall activation sequence is minimal as demonstrated in previous studies [34], [36].

The robustness of an imaging modality is crucial to its clinical performance. The imaging results from 13 patients in various age and gender are in good concordance with the procedural outcomes. The observations indicate that CESI is able to provide reliable imaging results in a realistic clinical setting and is robust against the disturbances common in clinical environments, such as movement artifact and the surrounding electronic interferences. The activation patterns from 12 of the 13 patients are also in agreement with the invasive mapping results. In the remaining case where discrepancies were found between the imaged and the intra-cardiac mapped activation patterns, the CESI imaging results are still in good agreement with the procedural outcomes.

As can be observed in the results, CESI is able to non-invasively image the cardiac electrical activities on a beat-to-beat basis in a clinical setting. The focal mechanisms of the ventricular arrhythmias in the current patient population are well captured in the imaged results. Although in this study, the ectopic beats were mainly isolated PVCs, the technique does not have intrinsic difficulties for imaging arrhythmias with continuous trains of ectopy such as sustainable or unsustainable VT trains, which are usually difficult to map in the EP lab. Much less demanding in medical expertise or effort to wait for or induce recurring arrhythmias to complete a sequential mapping, CESI can image the ectopic beats in their natural state, relatively free from the possible perturbation in the EP lab. The simple setup and whole heart imaging capability also enable CESI to perform as a long term monitoring modality where detailed information regarding arrhythmic mechanisms, especially for patients who have a changing dynamic, can be gathered to provide additional information for treatment or diagnosis purposes in various arrhythmia activities.

Previously, the CESI method has been validated via computer simulation and initial animal studies with simultaneous recordings by 3D transmural electrodes in rabbits. However, for clinical studies, the highly invasive approaches are not applicable in patients. Activation sequences and ablation sites on the endocardium mapped by invasive EP study can be utilized to assess the capability and performance of the method in a clinical environment. The present results have shown that CESI possesses the potential to function as a pre-procedural assessment and ablation planning tool with its full coverage and simple setup. Moreover, it can be a promising further investigation to evaluate the performance of the CESI technique as an imaging modality function in the EP lab in real time to the ablative procedures.

Despite the promising results aforementioned, the present study is limited by the nature that the BSPM mapping is conducted prior to the invasive mapping rather than simultaneously. Compared to a simultaneous study where all the geometry and ECG recordings, invasive and non-invasive, can be co-registered during the ablation, a pre-procedural study design would suffer from an extra error in registration, both spatially and temporally. The study is also limited by the diversity of arrhythmic morphologies with most PVCs initiating from the most common initiation site of RVOT. However, the signal from the regions with very low electrical activity, such as on the far end of RVOT and LVOT, will be very small in BSPM due to volume conduction. To avoid inaccurate reconstruction and provide accurate and robust imaging results, the tissue in the far end of RVOT and LVOT is not included in the present study. Future investigations should be conducted in a large patient population with diverse sites of arrhythmia initiations.

V. CONCLUSION

The present study reports, for the first time to our knowledge, the clinical performance of the noninvasive three-dimensional cardiac activation imaging technique in a group of 13 patients undergoing catheter ablation with ventricular arrhythmias. Quantitative comparisons have been made to the EP study results, ablation sites and ablation outcomes in all patients. Our results show that the non-invasive cardiac activation imaging results have very good concordance with the invasive mapping results and the ablation outcomes, and the technique promises to serve as a clinically useful tool to provide crucial arrhythmia information assisting the ablative procedure noninvasively.