Original Article. Experimental Validation of Noninvasive Epicardial and Endocardial Activation Imaging

Transcription

1 Original Article Experimental Validation of Noninvasive Epicardial and Endocardial Activation Imaging Peter Oosterhoff, PhD; Veronique M.F. Meijborg, PhD; Peter M. van Dam, PhD; Pascal F.H.M. van Dessel, MD, PhD; Charly N.W. Belterman, VA; Geert J. Streekstra, PhD; Jacques M.T. de Bakker, PhD; Ruben Coronel, MD, PhD*; Thom F. Oostendorp, PhD* Background Noninvasive imaging of cardiac activation before ablation of the arrhythmogenic substrate can reduce electrophysiological procedure duration and help choosing between an endocardial or epicardial approach. A noninvasive imaging technique was evaluated that estimates both endocardial and epicardial activation from body surface potential maps. We performed a study in isolated and in situ pig hearts, estimating activation from body surface potential maps during sinus rhythm and localizing endocardial and epicardial stimulation sites. Methods and Results From 3 Langendorff-perfused pig hearts, 180 intramural unipolar electrograms were recorded during sinus rhythm and ectopic activation, together with pseudo body surface potential map ECGs in 2 of them. From 4 other anesthetized pigs, 64-lead body surface potential maps were recorded during sinus rhythm and ventricular stimulation from 27 endocardial and epicardial sites. The ventricular activation pattern was computed from the recorded QRS complexes. For both Langendorff-perfused hearts, the calculated epicardial and endocardial activation patterns showed good qualitative correspondence to the patterns obtained with needle electrodes. Absolute timing difference for sinus rhythm was 10±5 and 11±8 ms respectively, and for ectopic activation 6±5 and 7±6 ms, respectively. Calculated activation for the in situ hearts in sinus rhythm was similar to patterns recorded in Langendorff-perfused hearts. During stimulation, the distance between the stimulation site and calculated site of earliest activation was 18 (15 27) mm, and 23 of 27 stimulation sites were correctly mapped to either endocardium or epicardium. Conclusions Noninvasive activation imaging is able to determine earliest ventricular activation and discriminate endocardial from epicardial origin of activation with clinically relevant accuracy. (Circ Arrhythm Electrophysiol. 2016;9:e DOI: /CIRCEP ) Key Words: electrocardiography imaging inverse electrocardiography mapping multimodality imaging noninvasive imaging Ventricular ablation therapy is often preceded by endocardial mapping procedures to detect the target sites for ablation. 1 The application of direct endocardial activation mapping is often hampered by the poor inducibility of arrhythmias in the sedated patient, poor hemodynamic tolerance of the induced arrhythmia, the risks associated with left ventricular catheterization, and the lack of epicardial activation information. 1,2 Noninvasive imaging of cardiac activation in the awake patient based on body surface electrocardiography (the inverse approach), therefore, is a potentially attractive method to guide ablation therapy and may help to increase its success rate. 3 5 Inverse calculation methods are based on either estimation of the epicardial electrogram (epicardial potential method) or on reconstruction of the current-generating activation wavefront (this article). The epicardial potential method uses body surface potentials to reconstruct the electrograms on the epicardium. Epicardial activation and repolarization times can be derived from these reconstructed electrograms, and under certain assumptions, some intramural excitation characteristics can be inferred before epicardial breakthrough. 6 8 This general approach was first suggested in the 1970s, and a variant of this method is used in the CardioInsight ECVUE system The inverse method used in the current study reconstructs the activation times on both the endocardial and epicardial surface, allowing discrimination of epicardial from endocardial foci. 14 We present results from a study in which endocardial and epicardial activation sequences, both for ventricular stimulated and for atrioventricular-conducted activation, were reconstructed from (pseudo) body surface potentials. The study consists of 2 parts. In part I, transmural activation mapping was Received November 26, 2015; accepted June 20, From the Donders Institute for Brain, Cognition & Behaviour, Radboud University Nijmegen Medical Center, Nijmegen, The Netherlands (P.O., P.M.v.D., T.F.O.); Department of Clinical and Experimental Cardiology (P.O., V.M.F.M., P.F.H.M.v.D., C.N.W.B., J.M.T.d.B., R.C.), and Department of Biomedical Engineering and Physics (G.J.S.), Academic Medical Center, Amsterdam, The Netherlands; ICIN-Netherlands Heart Institute, Utrecht, The Netherlands (V.M.F.M., J.M.T.d.B.); and IHU Institut de Rythmologie et Modélisation Cardiaque, Fondation Bordeaux Université, Bordeaux, Pessac, France (R.C.). *Drs Coronel and Oostendorp contributed equally to this work. Correspondence to Peter Oosterhoff, PhD, 126 Department of Cognitive Neuroscience, Donders Center for Neuroscience, Radboud University Medical Center, Geert Grooteplein 21, Nijmegen, The Netherlands. P.Oosterhoff@gmail.com 2016 American Heart Association, Inc. Circ Arrhythm Electrophysiol is available at DOI: /CIRCEP

2 2 Oosterhoff et al Noninvasive Imaging of Epi/Endocardial Activation WHAT IS KNOWN Preprocedural, noninvasive information about the endocardial or epicardial origin of ectopic activation is essential for the choice of the best approach for ablation of focal tachycardia. Inverse ECG methods using an equivalent dipole layer (EDL) source model can be used to estimate both endocardial and epicardial activation times from body surface potentials. WHAT THE STUDY ADDS This study validates the EDL inverse method for noninvasive estimation of endocardial and epicardial activation times from body surface potential maps recorded during sinus rhythm or ectopic activation in in situ structurally normal porcine hearts. These noninvasively estimated ectopic activation times allow reliable identification of the endocardial or epicardial source of ectopic activation. performed in Langendorff-perfused, isolated pig hearts, while recording pseudo body surface maps (BSMs) from the boundary of a fluid-filled container with integrated electrodes in which the heart was suspended. In part II, BSMs and catheter electrograms were recorded simultaneously in intact pigs. The data from part I were used to adapt our inverse method to porcine electrophysiology and to evaluate it using a simple, homogeneous volume conductor (ie, the fluid-filled container), with transmural activation mapping as a reference. However, BSMs are measured on the surface of the torso, which is irregularly shaped, and contains organs with low conductivity (eg, lungs) and blood with high conductivity. This will affect the current distribution within the thorax and, thereby, the potentials generated at the body surface. Therefore, experiments in intact adult pigs were used to show the ability of our noninvasive mapping technique to reliably determine ventricular activation and discriminate endocardial from epicardial sites of activation origin in situ. Methods Animal Handling All experiments were approved by the committee for experiments on animals of the Amsterdam Medical Center, and animal handling was in accordance with the Dutch Law on Animal Experimentation and the European Directive for the Protection of Vertebrate Animals Used for Experimental and Other Scientific Purposes (European Union Directive 86/609/EEC). The experiments were performed under general anesthesia. Premedication consisted of ketamine (15 20 mg/kg) and midazolam (0.8 mg/kg). Anesthesia was induced with sufentanil (50 μg) and midazolam (0.5 mg/kg) and maintained with isoflurane (1.5% in O 2 and air, 1:2), and for the in situ experiments, a continuous infusion was performed with sufentanil (5 10 μg/kg per hour), midazolam (1 2 mg/kg per hour), ketamine (10 mg/kg per hour), and pancuronium bromide (0.1 mg/kg per hour). For the Langendorff experiments (Part I), pigs (n=3, body weight kg) were anesthetized, and the heart was excised and immediately placed in ice-cold Tyrode s solution. Then blood was collected, and a 1:2 mixture of blood and Tyrode s solution was prepared. The aorta was cannulated and connected to the perfusion system. For the in situ experiments (Part II), anesthetized pigs (n=4, body weight kg) were fixed in a wooden frame to obtain a stable body position throughout the procedure. Cardiac and full-body computed tomography was performed, which was later used to obtain heart, lung, and torso geometry, as well as the location of the ECG electrodes. Part I: Data Collection and Analysis of Langendorff Experiments Three Langendorff-perfused pig hearts were used to explore porcine ventricular activation and validate our imaging method in a simple, homogeneous volume conductor. The apex of the heart was fixed to the bottom of the container to reduce movement. Endocardial and epicardial activation times were determined using the time of steepest downslope in unipolar electrograms collected from 45 transmural needles, each with 4 electrode terminals (interelectrode distance 4 mm) and impaled in the left and right ventricular wall and interventricular septum. In 2 experiments, the hearts were placed in a cylindrical container filled with perfusion fluid, whereas pseudo-bsms were collected from 61 electrodes on the internal surface of the container, referenced to mean of all recorded signals. The position of the heart within the container was reconstructed using multiple photographs of the heart inside the container after the fluid was drained. Locations of the needle electrodes within the heart were determined from the observed needle path after dissection of the heart. For more detailed methods, see Meijborg et al. 15 We analyzed recordings of sinus rhythm or atrial-paced, atrioventricular-conducted rhythm (both further referred to as sinus rhythm) and of spontaneous ventricular ectopic activations. Part II: In Situ Data Collection Sixty-four ECG electrodes were attached to the torso in a 10 6 configuration and to the front and hind legs (Figure 1A) before full-body and cardiac computed tomography scans were performed. Subsequently, introducer sheaths were placed in the jugular vein and carotid artery to provide access to the left and right ventricular endocardium and to the right atrium. An introducer sheath was placed in the pericardial space through a superior access. This allowed epicardial catheterization with minimal disturbance to the electric current distribution within the thorax and, thus, to the ECG. A screw-in stimulation electrode was placed in the lateral right endocardial atrial wall. Under fluoroscopic guidance, 2 steerable multipolar catheters were placed on the endocardial and epicardial surface, at opposite sides of the ventricular wall. Catheter electrode positions were recorded with fluoroscopy from at least 2 documented angles (BV Pulsera, Philips Medical Systems, The Netherlands). Unipolar catheter electrograms and ECG were recorded simultaneously (Biosemi, The Netherlands; sampling frequency 2048 Hz, bandwidth Hz) at sinus rhythm and during endocardial and epicardial stimulation. Bipolar sequential atrial and ventricular stimulation was used to avoid the occurrence of a retrograde P wave disturbing the paced QRS complex. The atrioventricular delay was set short enough to prevent fusion with the atrioventricular conducted activation. Ventricular stimulus amplitude was set just above the diastolic stimulation threshold. During recording, mechanical ventilation was temporarily interrupted to obviate the changes in the ECG because of alterations in thorax dimensions during respiration and to allow measurements independent of respiration cycle. Recording and ventricular stimulation was repeated at several sites on the left and right ventricle. Part II: Post Hoc Data Analysis in In Situ Experiments Triangulated meshes were constructed from computed tomography using custom software (GeomPeacs, Peacs, The Netherlands)

3 3 Oosterhoff et al Noninvasive Imaging of Epi/Endocardial Activation Figure 1. Computed tomography (CT) derived geometry. A, Geometry reconstructed from CT scans containing ventricles, blood cavities, lungs, thorax, and body surface potential map (BSM) electrode positions (circles). B, A cross section of the triangulated surface bounding the ventricular myocardium, showing the nodes and connecting triangles on the endocardium and epicardium. representing the surfaces of torso, lungs, atria, and ventricles, as well as ECG electrode positions (Figure 1). The triangulated surface bounding the ventricular myocardium comprised 1500 nodes (intersections of the mesh). Catheter electrode positions were reconstructed from fluoroscopy for each stimulation site (Figure 2). Images from different angles were annotated (ECG and catheter electrodes) and combined into a single 3D reconstruction, which was aligned to the computed tomography derived geometry using the ECG electrode positions. Reconstructed catheter positions were mapped to the appropriate ventricular surface (epicardium or endocardium). The method was tested with a phantom and had a positional error of 1.5 ( ) mm (median [25th 75th percentile]). For each ventricular stimulation site, 5 QRS complexes in the recorded BSMs were selected that showed atrial and ventricular capture. ECG components resulting from atrial depolarization and repolarization were removed by subtracting an ECG template constructed from BSMs recorded during conducted atrial-paced complexes. To limit noise, an average of the corrected 5 complexes was used for the noninvasive estimation of ventricular activation times. The site of earliest calculated activation was compared with the actual position of the stimulation catheter. Ventricular activation during sinus rhythm was calculated from the recorded BSMs in all 4 animals. A qualitative comparison was made to the overall pattern recorded with needle electrodes in the isolated hearts (Part I). For 2 animals, catheter electrograms were recorded during sinus rhythm (10 epicardial and 1 endocardial signal). Activation times were determined using the steepest point of the downslope as moment of activation. Data from different catheter positions were combined into a single activation map and compared with the calculated activation times in sinus rhythm from the same animal using absolute difference and Pearson s correlation coefficient. Relative difference and Pearson s correlation coefficient were used to quantify the match between calculated and recorded ECG. 16 Activation Imaging Method The activation imaging technique uses a forward model that, given a set of activation times, calculates the resulting body surface potentials that make up the QRS on the ECG. The QRS is the result of current generated Figure 2. Projecting catheter positions. To project stimulation sites and catheter-derived activation times on the geometry used by the inverse method, catheter positions must be aligned to the CT-derived geometry of the heart. A, Catheter electrode positions are reconstructed using fluoroscopy images from multiple angles. Shown here is left to right view. Visible body surface potential maps (BSM) electrodes and catheter electrodes are annotated (green markers). B, Reconstructed 3D catheter electrode positions (gray spheres) are aligned to the CT-derived geometry using the BSM electrodes visible on the fluoroscopy images (not shown) and projected on the nearest ventricular surface.

4 4 Oosterhoff et al Noninvasive Imaging of Epi/Endocardial Activation at the border between polarized and depolarized cells (ie, the activation wavefront). 17 The imaging technique uses an equivalent dipole layer (EDL) encompassing the surface of the ventricular myocardium (both endocardium and epicardium) that generates an identical extracardiac current as is generated by the activation wavefront The strength of this dipole layer is proportional to the local transmembrane potential at the ventricular surface and is, therefore, completely defined by the endocardial and epicardial activation times. The boundary element method is used to calculate the potentials on the body surface generated by the EDL current source; it models the passive volume conductor effects of the torso, assuming different isotropic conductivities for blood, heart muscle, and lungs. Consequently, this forward model transforms the ventricular activation times into simulated potentials on the body surface. Next, this forward model is used to solve the so-called inverse problem; determining a set of endocardial and epicardial activation times that generate a QRS that best fit the recorded QRS. This encompasses 2 steps: in the first step, an anisotropic propagation model is used to construct sets of activation patterns by subsequently considering each node of the ventricular mesh as a focus and calculating the resulting QRS complexes. 21 The activation pattern that results in calculated QRS waveforms that best match the recorded QRS is identified. Additional foci are added until no further improvement is obtained in the match between the measured and calculated QRS morphologies. The resulting activation pattern is used as an initial estimate in the second step. In the second step, iterative optimization (Levenberg Marquardt) is used to fine-tune the activation times at all nodes, thus, further improving the fit between calculated and measured ECG. The inverse problem of electrocardiography is ill-posed and, consequently, is very sensitive to input errors (eg, errors in the geometric model). Hence, a regularization term is introduced to guard against unphysiological solutions. The version used is the surface Laplacian on activation times, which suppresses excessive irregularities in activation patterns. 6,16,22 Further details of the method have been described in van Dam et al. 21 In the present study, some additional steps were implemented, required for the handling of the porcine heart morphology (extensive transmural Purkinje network). 23 The propagation model used in the initial estimate assumes an isotropic, low-conduction velocity of 0.4 m/s during the leading phase of the QRS (until the slope of the RMS curve reaches 10 mv/s) to represent slow propagation because of load mismatch close to the stimulation electrode. 24 For the remainder of the activation, anisotropic propagation is assumed, and the resulting activation times are scaled to fit the recorded QRS duration. Low transmural dispersion of activation times was previously reported in pig hearts and confirmed in our recordings in isolated hearts (see Results section). 15,25 However, low transmural dispersion will also augment the relative error introduced by the limited resolution of our triangulated ventricular geometry. To remove this effect from the initial estimate, we used an artificially high transmural conduction velocity set to 100 the velocity along the wall for the anisotropic part of the propagation model. Calculating the transmural dispersion that best fits the recorded QRS will now be handled by the iterative optimization step. Data Presentation and Statistics Data are presented as mean±standard deviation or median (25th 75th percentile). For evaluation of the percentage of correctly classified foci, a binomial test comparing to 50% (coin flip) was used. Proportions were compared with Fisher exact test. For all tests, a 2-sided P value was calculated, and a value of 0.05 was considered significant. Reported correlations are Pearson s coefficients. Statistical analysis was performed using the Matlab statistics toolbox (version 9.1 R2014b; The Mathworks Inc). Results Part I: Ventricular Activation During Sinus Rhythm in the Isolated Heart Ventricular activation maps were recorded in 3 isolated hearts during sinus rhythm. In 2 of these hearts, a pseudo-bsm was recorded as well. Figure 3A (blue tracing) shows an example QRS of the pseudo-bsms during sinus rhythm. The QRS vector is directed from basal-mid anterior to posterior base. The upper graphs of Figure 3B show the recorded epicardial and endocardial activation pattern, interpolated from the endocardial and epicardial needle electrode sites (white dots, 75 electrodes). 26 Note that this interpolation is less reliable in regions without needles, such as the right ventricular outflow tract. The transmural dispersion in the porcine myocardium was small. Average absolute difference between endocardial and epicardial activation time measured on the same needle in the 3 hearts was 2±3, 5±4, and 3±2 ms (respectively, 20, 12, and 23 needles with both endocardial and epicardial activation times). All 3 hearts showed earliest activation on the left endocardial septum or anterior wall, followed by a rapid activation of the anterior epicardium (<15 ms) and early breakthrough at the epicardial posterior wall (<20 ms). The latest activated regions were located at the right ventricular base of the free wall and the right ventricular outflow tract. Epicardial and endocardial activation times calculated from the recorded pseudo-bsms by the EDL inverse method in the 2 hearts showed the same patterns as described for the needle electrodes (example in Figure 3B, lower graphs). In the 2 hearts, the absolute difference between measured and calculated activation times for sinus rhythm was 10±5 and 11±8 ms at QRS widths of 56 and 65 ms for each heart, respectively (75 and 74 recording sites). The red trace in Figure 3A shows the near-perfect match of the calculated QRS to the measured one (blue trace, in this case, the correlation coefficient was 0.99, and relative difference was 0.12). Part I: Ectopic Activation in the Isolated Heart In the same 3 isolated hearts, transmural mapping of ventricular ectopic activation was performed: 2 spontaneous ectopic complexes originating from the left free wall and 1 paced complex from the high right septum. All 3 complexes showed centrifugal propagation from the site of origin. Absolute difference between endocardial and epicardial activation times on the same needle was 5±3, 4±4, and 5±4 ms for the 2 ectopic and the 1 paced complex, respectively (respectively, 19, 13, and 16 needles with both endocardial and epicardial activation times). Figure 4A shows the QRS of the pseudo-bsms and the calculated ECG for one of the complexes originating from the epicardial left free wall, together with the recorded and calculated activation times (Figure 4B). The site of origin for both spontaneously occurring complexes was correctly mapped to the endocardial or epicardial side of left free wall. The distance between measured and calculated site of first activation was 0 and 5 mm, respectively. For both complexes, the recorded activation times showed endocardial and epicardial activity within the first 5 ms, which may suggest an intramural site of origin. The absolute differences between measured and calculated activation time for the ectopic complexes were 6±5 and 7±6 ms (86 and 76 electrode sites, respectively). Part II: Calculating Activation Times During Sinus Rhythm in the In Situ Heart To evaluate the inverse calculation in situ, we recorded 64-channel BSMs during sinus rhythm in 4 intact pigs. As expected, the inverse calculation showed initial activation at

5 5 Oosterhoff et al Noninvasive Imaging of Epi/Endocardial Activation Figure 3. Sinus rhythm activation in an isolated heart. A, Overlay of recorded (blue tracing) and calculated (red tracing) pseudo body surface potential maps (BSM) QRS during sinus rhythm and atrioventricular conduction. Arrow shows the approximate QRS vector. B, Measured (top graphs) and calculated activation times (lower graphs) for the same beat as A, in superior, anterior, and posterior view. Note that the orifices of the aorta and the mitral valve are combined. The white dots indicate the recording needle positions. The measured activation times plotted were interpolated from the recorded values at the needles. the left endocardium at the septal or anterior wall and virtually simultaneous or early (<15 ms) breakthrough activation on the anterior epicardial myocardium. The activation pattern of the posterior epicardial myocardium appeared to be consistent with anterior to posterior activation. In 2 animals, a partial activation map was constructed using sequential electrogram recordings from multipolar catheters at several positions. Figure 5 shows the QRS complexes of a BSM recorded during sinus rhythm, together with the calculated activation map and the catheter-derived activation times (4 endocardial and 15 epicardial sites). The absolute differences between calculated and measured activation times were 11±5 ms (54 mainly posterior sites) and 5±3 ms (19 lateral and posterior sites; Figure 5) for the 2 catheter mappings,

6 6 Oosterhoff et al Noninvasive Imaging of Epi/Endocardial Activation Figure 4. Ventricular ectopic activation in an isolated heart. A, Overlay of recorded (blue tracing) and calculated (red tracing) pseudo body surface potential maps (BSM) QRS for spontaneous ectopic activation from the left free wall in an isolated heart. B, Measured (top graphs) and calculated activation times (lower graphs) for the same beat as A, in left, right, and superior view. Spheres indicate the measured (blue) and calculated (green) site of earliest activation. Note that the orifices of the aorta and the mitral valve are combined. respectively. Correlation coefficients for measured and calculated activation times were 0.53 and 0.82 (P<0.001 for both). Part II: Calculated Paced Activation Times in the In Situ Heart We recorded BSM during ventricular stimulation in 4 animals at a total of 27 stimulation sites (15 endocardial and 12 epicardial). Our inverse calculation correctly identified the endocardial or epicardial site of origin for 23 sites (23/27, 85%; P<0.001 versus 50%), and results were similar between the 4 animals (88%, 80%, 86%, and 86%, respectively). Four endocardial pacing sites were incorrectly mapped to the epicardium (2 left ventricle and 2 right ventricle). Results were similar for left and right ventricle (left ventricle 16/18 and right ventricle 7/9; P=0.58) and for endocardial and epicardial stimulation (11/15 versus 12/12; P=0.11). The distance between the site of earliest calculated

7 7 Oosterhoff et al Noninvasive Imaging of Epi/Endocardial Activation Figure 5. Example of sinus rhythm activation in situ. A, QRS of recorded body surface potential maps (BSM; blue trace) and calculated BSM (red trace). B, Calculated activation times in left lateral and superior view. Catheterderived activation times shown as colored spheres, using the same color scale. For reference, diameter of spheres is 3 mm, and distance from apex to septal base is 83 mm. Note that the orifices of the aorta and the mitral valve are combined. LFL indicates left front leg; LHL, left hind leg; RFL, right front leg; and RHL, right hind leg. activity and the actual position of the tip of the stimulation catheter was 18 (15 27) mm. In Figure 6, examples of QRS complexes and calculated activation times are shown for endocardial (Figure 6A) and epicardial stimulation (Figure 6B). Discussion In the current study, we compared endocardial and epicardial activation times calculated from body surface ECGs to invasively recorded data. This comparison was made during sinus rhythm and during focal activation, both in isolated and in situ pig hearts. Needle mapping during sinus rhythm showed a consistent activation pattern in all 3 isolated hearts, in terms of site of earliest activation, site of epicardial breakthrough, and general pattern. For the isolated hearts, calculated activation times from pseudo-bsm fitted closely to the measured activation times. For in situ hearts in sinus rhythm, the calculated activation times fitted well to the activation times obtained from catheter data and showed patterns similar to the needle mappings in isolated hearts. For focal activation, localization of the site of origin was excellent in the isolated hearts, where pseudo-bsm electrodes are placed at similar distances all around the heart in a container filled with high conductive perfusion fluid. These results were similar to the reports by others using fluid-filled tanks with realistic torso geometry. 7,27 Also in the in situ setting, we were able to reliably map ventricular paced activation. The EDL inverse was highly accurate (85%) in discriminating endocardial from epicardial origin of activation, despite the positional error (18 [15 27] mm) being larger than the average wall thickness ( 5 mm for the right ventricle and 10 mm for the left ventricle). This can be explained by the fact that the surface at which activation originates does not only affect the socalled initial vector of the ECG, but also changes the shape and direction of the transmural activation wavefront later in the QRS complex. These QRS changes are not easy to fit in general terms because they are the result of interaction between the location at which the activation originates and the geometry of the heart. The EDL method, using detailed geometry of the heart, is capable of capturing these QRS differences and calculates the most likely activation pattern. The ability to discriminate endocardial from epicardial foci demonstrates the potential clinical value of the EDL inverse method.

8 8 Oosterhoff et al Noninvasive Imaging of Epi/Endocardial Activation Figure 6. Examples of stimulated activation in situ. A, On the left, QRS of recorded body surface potential maps (BSM; blue trace) and calculated BSM (red trace) during endocardial stimulation. On the right, calculated activation times together with calculated endocardial focus (green sphere) and actual position of stimulation catheter (blue sphere) in superior view. Distance of calculated focus to stimulation site is 7 mm. B, As in A, for epicardial stimulation. Activation map in left to right view. Distance of calculated focus to stimulation site is 19 mm. For reference, diameter of spheres is 10 mm, and distance from apex to septal base is 82 mm. Note that the orifices of the aorta and the mitral valve are combined. LFL indicates left front leg; LHL, left hind leg; LV, left ventricle; RFL, right front leg; RHL, right hind leg; RV, right ventricle; and RVOT, right ventricular outflow tract. Equivalent Dipole Layer Inverse The approach of the EDL inverse applied in this article is to maximize the use of an a priori information about electrophysiology and geometry of the heart (both in the source model and in the regularization method) to estimate both endocardial and epicardial activation times. The source model consists of a (current) dipole layer located on the endocardial and epicardial surface of the ventricles; use of a precise, personalized geometry of heart generates realistic calculated body surface potentials, which will improve our solution. The strength of the EDL current source is proportional to the local transmembrane potential. Geselowitz showed that the EDL source at the myocardial surface is equivalent to the actual sources within the myocardium (ie, generates the same body surface potentials) if the anisotropy ratio of the intracellular medium is equal to that of the extracellular medium. 19,20 Because there is no clear-cut consensus about the values of the intra- and extracellular conductivities, equal anisotropy ratios in these spaces are assumed, and hence, the EDL source model represents an approximation However, Modre et al showed in a model study that anisotropy had a minor impact on the quality of the EDL-based inverse. 31 In the EDL model, the source strength is proportional to the transmembrane potential at the myocardial surface. In the present study, we have only regarded myocardial activation, and consequently, myocardial cells were assumed to be either at rest of completely depolarized. The source strength is, therefore, described as a step-like function,

9 9 Oosterhoff et al Noninvasive Imaging of Epi/Endocardial Activation defined by the local activation time, for which the amplitude in healthy tissue has been determined experimentally. 16,32,33 The EDL source model can readily be adapted to include a varying source strength, such as during repolarization or acute ischemia, as in fact is implemented in ECGSIM. 34 An EDL-based inverse method applied to both activation and repolarization would require a parameterized model of the whole action potential. When limited to activation, use of the EDL source model results in an inverse method that directly translates the full QRS complex from the measured body surface potentials to an endocardial and epicardial (but not intramural) cardiac activation map; in this way, it combines information from the spatial and from the temporal domain. Regularization is performed on activation times, allowing the use of more (a priori) knowledge on electrophysiology. In this study, we used the surface Laplacian on activation times, which is analogous to limiting the spatial variability of propagation velocity. This method of regularization also prevents unrealistic solutions, with an excessive number of transmural breakthrough sites. Results from the current study show that the EDL inverse is well capable of noninvasively estimating endocardial, septal, and epicardial isochrones, including the site of origin, in structurally normal hearts. For application in clinical settings, care has to be taken whether model assumptions and constraints are still valid when studying the pathological heart. The strength of regularization must still allow local slowing of propagation to detect regions with slowed activation (see Linnenbank et al for an application to sodium block in Brugada disease). 35 The EDL method can also be applied in patients with large transmural infarctions if accurate infarct location and complete morphology are included in the cardiac geometry. 36 The source model used in the potential inverse does not rely on any assumption or a priori knowledge on cardiac electrophysiological processes. Instead, various regularization methods have been proposed, which can be used to impose explicit constraints depending on the application (eg, zeroorder Tikhonov regularization, which limits the epicardial electrogram amplitude). 37,38 Although the method cannot reconstruct endocardial isochrones, the morphology of the epicardial electrogram does contain markers indicating a nonepicardial origin of activation. Taccardi et al showed that the depth of stimulation altered the morphology of the epicardial electrogram in structurally normal dog hearts; both the epicardial signal morphology opposite to the stimulation site and the rotation of flanking potential maxima reflected the depth of stimulation. However, the patterns for endocardial and intramural stimulation did partially overlap. 6 Discrimination between endocardial, intramural, and epicardial origin of activation might be improved by careful analysis of the signal morphology if the local myocardial fiber direction is known. Oster et al showed that these epicardial potential markers were also present in reconstructed electrograms, using an epicardial potential inverse method for dog hearts suspended in a torso tank. 7 In a clinical setting, Wang et al further validated this approach: in patients having ventricular tachycardia, the origin of activation was determined in an invasive electrophysiological procedure and compared with results from the epicardial potential inverse. 8 Consistent with findings from Taccardi et al, all 6 patients with epicardial origin showed a strictly negative reconstructed electrogram at the site of earliest activation, whereas an initial R wave was seen on the reconstructed epicardial electrogram for 7 out of 8 patients with endocardial or intramural sites of origin. Although requiring more details on patient geometry than the epicardial potential inverse, from a clinical perspective, the main advantage of the EDL method is the direct reconstruction of not only the origin, but also the complete set of epicardial and endocardial (including septal) activation times; the estimated origin of the activation is directly clear from the reconstructed activation map. Clinical Application of Noninvasive Activation Mapping Epicardial ablation of ventricular tachycardia is becoming more common with the development of minimally invasive techniques. Although being the initial choice in certain cardiomyopathies (eg, arrhythmogenic right ventricular cardiomyopathy), often the epicardial approach is attempted if endocardial ablation fails. 3 This leads to increased procedure time or multiple procedures. Moreover, both approaches differ considering preprocedural planning (eg, local versus complete anesthesia) and associated risk. Information of the location of the critical site in the awake patient will reduce procedure time considerably and may improve the success of the first ablation. Criteria for discriminating endocardial from epicardial site of origin using the standard 12-lead ECG have been proposed, but they often rely on subtle morphological QRS differences that may depend on the underlying pathology. 4,5 One solution may be to add information by using more than the standard 12 ECG leads, but this would require development of new, most likely more complex criteria. The EDL inverse method used in this article, applying automated analysis to make use of the full information contained in the 64 ECG leads, has the potential to significantly improve diagnosis. Study Setup: Rationale and Limitations In the current study, we combined measurements in isolated pig hearts with in situ measurement with closed thorax, combining left and right and endocardial and epicardial stimulation at several sites within the same animal. This animal model allowed us to perform extensive mapping in structurally normal hearts. The minimally invasive procedure used for epicardial catheterization enabled us to record realistic epicardial and full-body surface electrograms during the procedure. Furthermore, care was taken to determine the absolute position of the stimulation and mapping catheters using fluoroscopy. The difference between human and porcine electrophysiology potentially, however, may limit our results. We found that transmural dispersion in activation, both during sinus rhythm and ventricular stimulation, was much smaller in the pig heart than the values reported for humans. 39 In that respect, the pig may be a worst case model for discriminating endocardial from epicardial activation, and more robust discrimination is to be expected in man, where the ventricular walls are thicker, and transmural activation delay is much larger.

10 10 Oosterhoff et al Noninvasive Imaging of Epi/Endocardial Activation Conclusions We validated a noninvasive ECG inverse solution that can correctly map endocardial and epicardial activation and determine the site of origin of ventricular activation, whether endocardial or epicardial, with clinically relevant accuracy. This method can be applied in the awake patient and may guide clinical decision-making for ablation of ectopic activation, in particular where it concerns decisions on an initial epicardialor endocardial-oriented approach. Sources of Funding This study was supported by the Dutch Technology Foundation STW grant None. Disclosures References 1. Wilber DJ, Packer DL, Stevenson WG, eds. Catheter Ablation of Cardiac Arrhythmias: Basic Concepts and Clinical Applications. 3rd ed. Oxford, UK: Blackwell Publishing; Kaltenbrunner W, Cardinal R, Dubuc M, Shenasa M, Nadeau R, Tremblay G, Vermeulen M, Savard P, Pagé PL. Epicardial and endocardial mapping of ventricular tachycardia in patients with myocardial infarction. Is the origin of the tachycardia always subendocardially localized? Circulation. 1991;84: Della Bella P, Brugada J, Zeppenfeld K, Merino J, Neuzil P, Maury P, Maccabelli G, Vergara P, Baratto F, Berruezo A, Wijnmaalen AP. Epicardial ablation for ventricular tachycardia: a European multicenter study. Circ Arrhythm Electrophysiol. 2011;4: doi: / CIRCEP Vallès E, Bazan V, Marchlinski FE. ECG criteria to identify epicardial ventricular tachycardia in nonischemic cardiomyopathy. Circ Arrhythm Electrophysiol. 2010;3: doi: /CIRCEP Bazan V, Gerstenfeld EP, Garcia FC, Bala R, Rivas N, Dixit S, Zado E, Callans DJ, Marchlinski FE. Site-specific twelve-lead ECG features to identify an epicardial origin for left ventricular tachycardia in the absence of myocardial infarction. Heart Rhythm. 2007;4: doi: /j.hrthm Taccardi B, Macchi E, Lux RL, Ershler PR, Spaggiari S, Baruffi S, Vyhmeister Y. Effect of myocardial fiber direction on epicardial potentials. Circulation. 1994;90: Oster HS, Taccardi B, Lux RL, Ershler PR, Rudy Y. Electrocardiographic imaging: Noninvasive characterization of intramural myocardial activation from inverse-reconstructed epicardial potentials and electrograms. Circulation. 1998;97: Wang Y, Cuculich PS, Zhang J, Desouza KA, Vijayakumar R, Chen J, Faddis MN, Lindsay BD, Smith TW, Rudy Y. Noninvasive electroanatomic mapping of human ventricular arrhythmias with electrocardiographic imaging. Sci Transl Med. 2011;3:98ra84. doi: / scitranslmed Martin RO, Pilkington TC. Unconstrained inverse electrocardiography: epicardial potentials. IEEE Trans Biomed Eng. 1972;19: doi: /TBME Barr RC, Spach MS. Inverse calculation of QRS-T epicardial potentials from body surface potential distributions for normal and ectopic beats in the intact dog. Circ Res. 1978;42: Rudy Y, Burnes JE. Noninvasive electrocardiographic imaging. Ann Noninvasive Electrocardiol. 1999;4: Ramanathan C, Ghanem RN, Jia P, Ryu K, Rudy Y. Noninvasive electrocardiographic imaging for cardiac electrophysiology and arrhythmia. Nat Med. 2004;10: doi: /nm Haissaguerre M, Hocini M, Shah AJ, Derval N, Sacher F, Jais P, Dubois R. Noninvasive panoramic mapping of human atrial fibrillation mechanisms: a feasibility report. J Cardiovasc Electrophysiol. 2013;24: doi: /jce van Oosterom A. A comparison of electrocardiographic imaging based on two source types. Europace. 2014;16:iv120 iv Meijborg VM, Conrath CE, Opthof T, Belterman CN, de Bakker JM, Coronel R. Electrocardiographic T wave and its relation with ventricular repolarization along major anatomical axes. Circ Arrhythm Electrophysiol. 2014;7: doi: /CIRCEP Huiskamp G, Van Oosterom A. The depolarization sequence of the human heart surface computed from measured body surface potentials. IEEE Trans Biomed Eng. 1988;35: doi: / Wilson FN, Macleod AG, Barker PS. The distribution of the action currents produced by heart muscle and other excitable tissues immersed in extensive conducting media. J Gen Physiol. 1933;16: Cuppen JJ, van Oosterom A. Model studies with the inversely calculated isochrones of ventricular depolarization. IEEE Trans Biomed Eng. 1984;31: doi: /TBME Geselowitz DB. On the theory of the electrocardiogram. Proc IEEE Inst Electr Electron Eng. 1989;77: Geselowitz DB. Description of cardiac sources in anisotropic cardiac muscle. Application of bidomain model. J Electrocardiol. 1992;25(suppl): van Dam PM, Oostendorp TF, Linnenbank AC, van Oosterom A. Noninvasive imaging of cardiac activation and recovery. Ann Biomed Eng. 2009;37: doi: /s Huiskamp G. Difference formulas for the surface Laplacian on a triangulated surface. J Comput Phys. 1991;95: Glomset D, Glomset A. A morphologic study of the cardiac conduction system in ungulates, dog, and man: Part II: The purkinje system. Am Heart J. 1940;20: Kléber AG, Rudy Y. Basic mechanisms of cardiac impulse propagation and associated arrhythmias. Physiol Rev. 2004;84: doi: / physrev Allison JS, Qin H, Dosdall DJ, Huang J, Newton JC, Allred JD, Smith WM, Ideker RE. The transmural activation sequence in porcine and canine left ventricle is markedly different during long-duration ventricular fibrillation. J Cardiovasc Electrophysiol. 2007;18: doi: /j x. 26. Oostendorp TF, van Oosterom A, Huiskamp G. Interpolation on a Triangulated 3d Surface. J Comput Phys. 1989;80: Oster HS, Taccardi B, Lux RL, Ershler PR, Rudy Y. Noninvasive electrocardiographic imaging: reconstruction of epicardial potentials, electrograms, and isochrones and localization of single and multiple electrocardiac events. Circulation. 1997;96: Rush S, Abildskov JA, McFeer. Resistivity of body tissues at low frequencies. Circ Res. 1963;12: Roth BJ. Electrical conductivity values used with the bidomain model of cardiac tissue. IEEE Trans Biomed Eng. 1997;44: doi: / Keller DU, Weber FM, Seemann G, Dössel O. Ranking the influence of tissue conductivities on forward-calculated ECGs. IEEE Trans Biomed Eng. 2010;57: doi: /TBME Modre R, Seger M, Fischer G, Hintermüller C, Hayn D, Pfeifer B, Hanser F, Schreier G, Tilg B. Cardiac anisotropy: is it negligible regarding noninvasive activation time imaging? IEEE Trans Biomed Eng. 2006;53: doi: /TBME van Oosterom A, van Dam RT. Potential distribution in the left ventricular wall during depolarization. Adv Cardiol. 1976;16: van Oosterom A. Macroscopic source descriptions. In: Macfarlane PW, van Oosterom A, Janse M, Kligfield P, Camm AJ, Pahlm O, eds. Basic Electrocardiology. London: Springer-Verlag; 2012: van Oosterom A, Oostendorp TF. ECGSIM: an interactive tool for studying the genesis of QRST waveforms. Heart. 2004;90: Linnenbank AC, van Oosterom A, Oostendorp T, van Dessel P, Van Rossum AC, Coronel R, Tan HL, de Bakker JMT. Non-invasive imaging of activation times during drug induced conduction changes. In: Magjarevic R, Nagel JH, eds. IFMBE Proceedings. Berlin: Springer; 2007: Oostendorp T, Nenonen J, Korhonen P. Noninvasive determination of the activation sequence of the heart: application to patients with previous myocardial infarctions. J Electrocardiol. 2002;35(suppl): doi: /jelc Ghosh S, Rudy Y. Accuracy of quadratic versus linear interpolation in noninvasive Electrocardiographic Imaging (ECGI). Ann Biomed Eng. 2005;33: doi: /s x. 38. Milanič M, Jazbinšek V, Macleod RS, Brooks DH, Hren R. Assessment of regularization techniques for electrocardiographic imaging. J Electrocardiol. 2014;47: doi: /j.jelectrocard Durrer D, van Dam RT, Freud GE, Janse MJ, Meijler FL, Arzbaecher RC. Total excitation of the isolated human heart. Circulation. 1970;41: