Author + information
- Received February 28, 2004
- Revision received August 14, 2004
- Accepted August 23, 2004
- Published online December 7, 2004.
- Vivek Y. Reddy, MD*,* (, )
- Zachary J. Malchano*,
- Godtfred Holmvang, MD†,
- Ehud J. Schmidt, PhD‡,
- Andre d'Avila, MD*,
- Christopher Houghtaling, BS, MS*,
- Raymond C. Chan, PhD§ and
- Jeremy N. Ruskin, MD*
- ↵*Reprint requests and correspondence:
Dr. Vivek Y. Reddy, Cardiac Arrhythmia Service, Massachusetts General Hospital, 55 Fruit Street, Gray-Bigelow 109, Boston, Massachusetts 02114
Objectives In a series of in vitro and in vivo experiments, we evaluated the feasibility of integrating three-dimensional (3D) magnetic resonance imaging (MRI) and electroanatomic mapping (EAM) data to guide real-time left ventricular (LV) catheter manipulation.
Background Substrate-based catheter ablation of post-myocardial infarction ventricular tachycardia requires delineation of the scarred myocardium, typically using an EAM system. Cardiac MRI might facilitate this procedure by localizing this myocardial scar.
Methods A custom program was employed to integrate 3D MRI datasets with real-time EAM. Initially, a plastic model of the LV was used to determine the optimal alignment/registration strategy. To determine the in vivo accuracy of the registration process, ablation lesions were directed at iatrogenic MRI-visible “targets” (iron oxide injections) within normal porcine LVs (n = 5). Finally, this image integration strategy was assessed in a porcine infarction model (n = 6) by targeting ablation lesions to the scar border.
Results The in vitro experiments revealed that registration of the LV alone results in inaccurate alignment due primarily to rotation along the chamber's long axis. Inclusion of the aorta in the registration process rectified this error. In the iron oxide injection experiments, the ablation lesions were 1.8 ± 0.5 mm from the targets. In the porcine infarct model, the catheter could be reliably navigated to the mitral valve annulus, and the ablation lesions were uniformly situated at the scar borders.
Conclusions These data suggest that registration of pre-acquired magnetic resonance images with real-time mapping is sufficiently accurate to guide LV catheter manipulation in a reliable and clinically relevant manner.
A recent significant advance in the management of ventricular tachycardia (VT) has been the recognition that substrate-based catheter ablation can be employed to eliminate most scar-related VTs regardless of their hemodynamic effect (1–4). This strategy necessitates the use of an electroanatomic mapping (EAM) system to define the geometry of the ventricular chamber during sinus rhythm and then carefully delineate the location/extent of the myocardial scar on the basis of the electrogram characteristics (5–7). However, the fidelity of these three-dimensional (3D) electroanatomic substrate maps is dependent on the operator experience and skill, and even when performed by experienced operators, can still be a time-consuming and technically challenging procedure.
Because of its ability to provide detailed anatomic and physiologic information about normal and damaged myocardial tissue, cardiac magnetic resonance imaging (MRI) could greatly facilitate electrophysiology procedures. Magnetic resonance imaging is noninvasive, non-ionizing, and can generate images of high spatial resolution. Of particular relevance to catheter ablation of VT, delayed-enhancement contrast MRI has recently been developed to distinguish normal from chronically infarcted cardiac tissue with millimeter spatial resolution (8–10). The MRI enhancement observed relatively quickly (within tens of seconds) after injection of the MRI contrast agent, gadolinium, is related to vascular perfusion. However, delayed enhancement (defined as appearing >5 min after bolus injection) is selectively observed in scar tissue because of the relatively larger extracellular space, and therefore larger volume of distribution of gadolinium within this tissue. This technique has already been applied clinically to define the scar morphology in post-myocardial infarction (MI) patients (9).
In patients who are to undergo substrate-based catheter ablation, a pre-procedural cardiac MRI could serve as a useful “road-map” to guide the procedure. In the optimal scenario, the 3D cardiac magnetic resonance (MR) images would be integrated with the EAM information so as to guide catheter navigation to the infarct borders in a real-time fashion. To accomplish this, the pre-acquired 3D MRI dataset must be properly registered with the EAM system; that is, both the MRI and electroanatomic constructs must be properly aligned. To determine the feasibility of this catheter mapping paradigm, a series of experiments were performed using: 1) a plastic life-size reproduction of the left ventricle (LV) and arterial vascular system, 2) a normal porcine model with putative “targets” for catheter ablation to determine the degree of accuracy in the registration process, and 3) a porcine model of healed anterior wall MI to evaluate MRI-based catheter navigation to the borders of the myocardial scar. These experiments test the general hypothesis that registration of pre-acquired cardiac images with real-time EAM is sufficiently accurate to guide catheter navigation in the LV in a reliable and clinically relevant manner.
This protocol was approved by the Massachusetts General Hospital Subcommittee of Research Animal Care, and was performed according to institutional guidelines. The animal experiments were performed using a total of five normal pigs and six pigs with chronic anterior wall MI.
Aorta–LV phantom experiments
In vitro experiments were performed using a plastic model of the LV and aorta (Angiogram Sam, Medical Plastic Laboratory, Inc., Gatesville, Texas) (Figs. 1A to 1C).Initially, this phantom was filled with a dilute solution of gadolinium and then imaged by MRI. The MR images were segmented (Table 1)to obtain accurate 3D datasets of the LV and aorta. Real-time magnetic EAM was then performed as described subsequently to acquire endoluminal “aortic” points and endocavitary “ventricular points” using a retrograde aortic approach. A total of three separately complete EAM acquisitions of the aorta and LV were obtained. Five simulations were performed using each of these three datasets to study various registration strategies. At the start of each simulation, the points in the dataset were randomized and then sequentially added to mimic real-time EAM. This randomization ensured that the order of EAM point acquisition had minimal impact on the sequential registration process.
Overview of in vivo experiments
The in vivo porcine experiments were divided into two phases. In phase 1 (Fig. 2),normal animals initially underwent an iron oxide injection procedure to place “targets” for catheter ablation. The MRI was then performed to define the chamber geometry and to identify the location of the injections. During a subsequent electrophysiologic study, the electroanatomic and MRI datasets were registered; based on this registered MRI dataset, radiofrequency ablation lesions were applied as close as possible to the iron oxide injection “targets.” In phase 2, the porcine infarction models underwent MRI to again define the chamber geometry and to delineate the location of the myocardial scar. After registration of the electroanatomic and MRI datasets during a subsequent electrophysiologic study, the registered MRI dataset was used to guide the placement of radiofrequency ablation lesions at the borders of the scar.
Porcine infarct generation
As previously described, a closed-chested anterior wall MI procedure was performed in 25- to 35-kg pigs (5,6). Briefly, after an overnight fast, the animals were intubated and general anesthesia was maintained with 1.5% to 2.5% isoflurane. Using a femoral arterial approach, an angioplasty balloon was advanced to the mid-left anterior descending artery, and 60 to 80 μl of agarose beads (contour 75 to 150 μm emboli; Boston Scientific, Natick, Massachusetts) were injected. After recovery, the animals were housed in an animal facility for a minimum of three months to allow maturation into chronic myocardial infarcts, followed by subsequent MRI and electrophysiologic study.
The in vivo electrophysiologic studies were performed by one of two operators on both: 1) normal pigs weighing 35 to 55 kg, and 2) anterior wall MI porcine models. Under general anesthesia, femoral vascular access was achieved and magnetic EAM of the aorta and LV was performed. As previously described, this system (CARTO; Biosense-Webster, Inc., Diamond Bar, California) allows for precise 3D mapping using a low-intensity magnetic field which can localize the mapping catheter with six degrees of freedom relative to a reference catheter: x-, y-, and z-positions in space and the roll, pitch, and yaw of the catheter tip (11). The accuracy of the system has been estimated at 0.8 mm and 5°.
Using a retrograde aortic approach, a 4-mm-tip mapping catheter (Navistar; Biosense-Webster, Inc.) was used to fully map the LV endocardium during sinus rhythm to achieve a fill threshold ≤15 mm. The aorta was mapped using the “vessel pullback” function—points are rapidly acquired by the system while withdrawing the catheter from an aortic sinus of valsalva to the descending aorta. Points were acquired in all three cusps of the aortic valve, and the descending aorta was rendered to the level of the inferior border of the heart shadow in the antero-posterior fluoroscopic projection. The electrograms are sampled at 1,000 Hz and recorded after bandpass filtering from 10 to 400 Hz. All points were acquired while gating to the peak of the surface QRS complex, approximating end-diastole.
In studies involving the porcine infarct models, the electroanatomic map was displayed as a naked geometric shell without any superimposed electrical information during the electrophysiologic study. Using the registered MRI dataset for catheter guidance, radiofrequency ablation lesions were applied in a temperature-controlled mode limit of 55°C with up to 50 W for 30 to 45 s. During retrospective post-processing analyses, electroanatomic identification of the infarcted myocardium was delineated by bipolar voltage amplitude criteria; scar is defined as bipolar voltage amplitude ≤1.5 mV, and a color scale range of 0.5 to 1.5 mV was displayed (5–7).
The animals were placed on the gantry for MRI in a similar position as during the EAM procedure to minimize distortions in the shape and curvature of the torso. In the series of normal porcine experiments, a thermoplastic immobilization system to prevent movement of the upper torso (SecureFoam; Bionix, Inc., Toledo, Ohio) was additionally employed. These animals underwent MRI and EAM on the same day, whereas the porcine infarct models were imaged 2.4 ± 3.7 days (range 0 to 9 days) before EAM. The animals were mechanically ventilated, and the images were acquired at end-expiration. All MR images were acquired in a 1.5-T GE CV/I scanner (GE Medical Systems Inc., Waukesha, Wisconsin) equipped with a surface cardiac array coil. A breath-held 3D contrast-enhanced MR angiogram was used to image the descending aorta (repetition time [TR]/echo time [TE]/∞ = 6.6 ms/2.4 ms/45°, field of view [FOV] = 28 × 28 cm, 192 × 256 matrix, 2.2 mm slice width (sw), 32 slices/scan, 1 number of excitations [NEX]; 0.44 cc/kg gadolinium). Short-axis and long-axis cardiac images were acquired with sequential, breath-held two-dimensional SHARK-FEISTA (GE Medical Systems Inc.) (TR/TE/∞ = 6.3 ms/1.9 ms/50°, 8 to 12 views per segment, FOV = 26 × 26 cm, 256 × 224 matrix, 4 to 5 mm sw, 1 to 2 NEX, 20 phases/R to R interval, electrocardiographic gating). The location and extent of the infarcted tissue in the six animals with healed MI was determined by two-dimensional myocardial delayed enhancement (TR/TE/theta = 5.7 ms/1.5 ms/20°, 12 to 16 views per segment, inversion time = 90 to 170 ms, FOV = 26 × 26 cm, 256 × 224 matrix, 4 to 5 mm sw, 2 to 3 NEX, acquired 15 to 25 min post injection of gadolinium-diethylene triamine pentacetic acid) (8–10) and a modified two-dimensional double-inversion recovery fast spin echo (TR/TE = 3 R to R interval [2,500 to 3,500 ms]/144 ms, echo train length = 32, FOV 26 × 26 cm, 256 × 224 matrix, 4 to 5 mm sw, 1 to 2 NEX). The prescribed post-QRS delays were selected to coincide as best possible with late diastole in order to match the cardiac phase during EAM.
Injection of iron oxide
The injections were performed using an 8 French mapping catheter with a 27-gauge retractable injection needle (Noga-Star; Biosense-Webster, Inc.) advanced into the LV using a retrograde aortic approach (12). At one to three discrete sites within the LV, the needle was extended 4 to 6 mm into the myocardium and a total of 0.2 ml of iron oxide particles (0.4 mg/ml Feridex; Advanced Magnetics, Inc., Cambridge, Massachusetts) were injected per site. These injections were placed at a minimum of 20 mm apart from each other. Injections were never placed at the LV apex because of the relatively characteristic reproducibility of this location. To facilitate gross pathologic visualization of the injection site, either methylene blue or India ink was included with the iron oxide particles in three and two animals, respectively.
Before each real-time experiment, the MRI datasets were manually segmented to define the endocardium, aorta, and (in the porcine infarct models) the scarred myocardium. Segmentation was performed using customized Matlab software (The MathWorks, Inc., Natick, Massachusetts). For each anatomic structure of interest, the contours were manually segmented. These contours retain their spatial position in the subject-based coordinate system, which is defined at the start of the MRI acquisition according to the Digital Imaging and Communications in Medicine (DICOM) standard. By using this coordinate framework, structures segmented in different MRI series remain in the correct relative positions to one another. To improve the accuracy of the representations, the long-axis and short-axis images were typically both segmented for the aorta, endocardium, and scar. Software was written to allow for the transmission of real-time catheter location via serial communication from the EAM system to a separate registration computer. Custom software was written using C++ and the Visualization toolkit to receive the EAM data and perform the registration algorithms and visualization (13). This custom program: 1) displays and allows one to electronically manipulate the segmented MRI dataset; 2) re-creates the electroanatomic map based upon the transmitted spatial coordinates and electrogram information (including a simulation of the interpolations of the electrogram data performed by the EAM system); 3) registers the two datasets based upon selected features of each dataset (for example, registration based solely on the aorta of the MRI with the mapping catheter “pull-backs” in the aorta); and 4) allows real-time visualization of the catheter tip within the registered MRI anatomic framework.
At the beginning of each in vivo procedure, the contours from the segmented MR datasets were loaded into the registration software. Surfaces representing the aorta, the LV endocardium, and in the relevant cases, the myocardial scar, were generated for electronic display and manipulation. The registration process then consisted of up to three phases. First, an initial coordinate transformation matrix between the two coordinate systems (EAM-based table coordinates and MRI coordinates) was calculated. This provides a rough estimation of the anterior-posterior and superior-inferior orientations. Second, the points acquired in the aorta (acquired using a total of 4 pullbacks from the aortic valve cusps to the descending aorta) were employed as an internal fiducial structure to estimate the transformation between the two imaging modalities: registration of the EAM point-to-MR-surface strategy was performed by allowing the iterative closest points (ICP) algorithm to converge upon this transformation (14). This algorithm attempts to minimize the mean distance from each EAM point to the MRI surface. Once a satisfactory initial pose was found, the registration was improved incrementally by the addition of points within the LV endocardium. Upon receiving the spatial coordinates of each newly acquired EAM point, the registration computer re-evaluates the current registration by calculating the root-mean-squared distance for each EAM point to the closest point on the MR surface. In the FeO injection and porcine infarct in vivo experiments, this registration phase of the procedure continued typically until there was minimal discernable relative movement of the two datasets with the inclusion of each successive point in the incremental registration process.
At the conclusion of each mapping procedure, the animal was killed and the heart was immediately explanted and examined. In normal animals, the distance from the center of the ablation lesions to the endocardial projection of the iron oxide injection “targets” was measured. The fact that the ablation catheter was not always maneuverable to precisely over the target injection site was factored into determining this error. For example, if the ablation point was placed 2 mm septal to the registered MRI-based injection site during the electrophysiologic study, and the lesion was visualized 5 mm septal to the injection site upon pathologic evaluation, the final error would be defined as 3 mm. In the infarcted animals, the location of the lesions was noted in relation to the borders of the scar.
Registration of the LV in a phantom model
Using the plastic phantom model of the LV and aorta, real-time electrophysiologic studies were performed to evaluate two registration strategies: registration using endocardial LV points only, and registration using both endoluminal aorta points and the LV points. In the first strategy, after the initial coordinate transformation, endocavitary LV points were sequentially registered to the MRI-LV dataset. When the mean distance of the complete EAM LV point dataset-MRI surface was used to evaluate the efficiency of registration, proper registration appeared to be achieved after acquiring only ∼10 endocardial points. To determine whether this represented ground truth or a local minimum solution, the registration accuracy was reassessed based upon the mean distance of the aorta points; that is, registration was based solely upon the LV points, and the accuracy based upon the EAM aorta point-MRI surface distance. As shown by the misalignment of the aorta in Figure 1D, it is clear that use of the LV alone for registration resulted in an imperfect local minimum solution.
To improve registration, the strategy was modified: after the coordinate transformation, the aorta points were registered to the 3D MRI-aorta dataset. In this scenario, excellent registration was achieved after incorporating only the first three endocardial LV points into the registration process; the mean distance of the EAM LV point-MRI surface was ∼1.5 mm (Fig. 1E). When 10 to 20 additional LV points were nonetheless incorporated in the registration process, the registration did appear to improve by ∼0.3 mm. Thus, employing the endoluminal aorta points in the registration process greatly improved the alignment between the MRI and EAM data.
Feasibility and accuracy of in vivo LV registration
The two registration strategies were then compared in the normal porcine models. As seen with the phantom experiments, when only the LV points were used in the registration process (Fig. 3A),there was a tendency for the registration to result in local minimum solutions. With the addition of all the LV points, there was a tendency to improve registration; however, in one porcine experiment, even with use of >60 LV points, registration resulted in a local minimum that could not be improved. On the other hand, when the aorta points were employed in the registration process (Fig. 3B), the mean distance of the EAM LV point-MRI LV surface was only ∼4 mm. Furthermore, with the incorporation of ∼20 LV points in the registration process, this improved to ∼3.5 mm, and additional points did not further improve the registration.
To independently establish the accuracy of the registration process, the LV was registered based upon the aorta points and 20 endocardial points, and then the mapping catheter was maneuvered to the mitral valve annulus based upon the registration MR image. As shown in Figure 4,the catheter tip was invariably localized to the corresponding annular position as defined by electrophysiologic criteria. Based upon the MRI, the catheter could also be guided to the LV apex (not shown). As a second confirmatory measure of the registration, iron oxide injection sites were targeted for radiofrequency ablation (Fig. 5,Table 2).Based upon the gross pathologic analysis, the mean error of the ablations from the “targets” was only 1.8 ± 0.5 mm (range 0 to 3.5 mm; n = 12).
MR-based catheter ablation at the porcine ventricular scar border zone
The MRI of the porcine chronic infarct models was performed 119 ± 12 days (range 102 to 133 days) after the infarction procedure. As with the normal porcine models, many of the registered EAM points were located outside the MRI endocardial surface. Also, the LV endocardial volume of the infarcted animals was smaller as determined by MRI than by EAM, 153 ± 60 cc versus 178 ± 28 cc, respectively. To register the datasets, a coordinate transformation was followed by registration of the aorta and 20 to 30 endocardial LV points. Once registered, the mean distance between the EAM-LV points and the MRI-determined LV was 4.9 ± 1.8 mm. Despite the volume difference between the two datasets, when radiofrequency lesions were targeted based solely upon the registered MR image, the lesions were uniformly situated at the scar borders (Fig. 6).
This study assessed the feasibility of aligning pre-procedural 3D MRI scans of the LV with intra-procedural EAM so as to guide LV catheter mapping and ablation. The major findings of this study include: 1) because the LV is somewhat symmetric along its long axis, registration of this chamber using endocavitary ventricular points alone (derived by EAM) is complicated enhances the accuracy of the registration process by limiting the rotation of the LV along its long axis; 3) significant volume differences can exist between the MRI and EAM datasets; and 4) despite these chamber differences, the registration is precise enough to guide catheter ablation to the border zones of a myocardial scar.
The concept of integrating pre-acquired MRI data into a surgical workstation to guide interventional procedures is not new to medicine. The idea of image-guided procedures was first exploited by neurosurgeons even before the advent of MRI or computed tomographic (CT) imaging technologies (15). However, these initial strategies had limited clinical utility because they were based upon standard anatomic models, and stereotaxy-based human surgery was confounded by the inherent anatomic spatial variability present between individuals. With the introduction of CT and MRI in medicine, patient-specific, image-guided volumetric stereotactic procedures are now standard techniques performed by surgeons to manage neuropathologic conditions such as intracranial neoplasia and movement disorders (15).
The application of this strategy to cardiovascular procedures poses challenges different from those encountered in neurosurgery. In the brain, accurate registration can be obtained before skull decompression by either utilizing external fiducial markers placed around the head, or by incorporating the unique topographic features of the human face (15). In the heart, however, cardiac structural features and position are affected by both the heart's contractile state and its position within the chest during the respiratory cycle. And unlike the topographically unique face, the chest is relatively featureless. On the other hand, the level of accuracy required in the registration process to guide cardiac catheter ablation procedures may not be so high. That is, a typical ablation lesion made by a saline-irrigated radiofrequency ablation catheter may be 5 to 10 mm in diameter. Thus, for clinically relevant proper alignment of pre-acquired cardiac MR images with intra-procedural cardiac mapping data, the registration strategy must negotiate the respiratory/cardiac motion of the heart, albeit with the understanding that the level of accuracy need not necessarily be as high.
Cardiac EAM identifies a number of endocardial surface points, whereas MRI generates an endocardial surface. Accordingly, the ICP algorithm was utilized to minimize the mean distance between the EAM points and the segmented MRI surface. However, whereas LV MRI produces high-resolution datasets, EAM typically consists of a modest number of points—up to a few hundred in the most detailed maps. If enough electroanatomic points are acquired, it is likely that even in a relatively smooth, minimally featured dataset such as the LV, the ICP algorithm would be capable of avoiding local minima and instead find a “global minimum” representing perfect registration. But in the ideal scenario, one would be able to register the two datasets rapidly without requiring a large number of electroanatomic points. However, given a limited number of electroanatomic points, the registration could result in a “local minimum” of poor overall alignment between the two datasets. Indeed, in the simulation experiments with the plastic heart model, registration based solely upon endocavitary LV points invariably resulted in a local minimum solution (Fig. 3A). On the other hand, these simulation experiments revealed that by incorporating an initial registration step to align the aorta, only a minimal number of endocardial LV points were required to properly register this chamber. This improved registration was likely related to the unique curvature of the aorta.
For the in vivo situation, cardiac motion was accounted for by acquiring both the EAM points and the MR images at end-diastole and at end-expiration (i.e., the level of the functional residual volume). In the ideal scenario, when properly gated and registered the LV MRI and EAM datasets would be identical—that is, the two imaging technologies would delimit an identical chamber shape and size. However, the in vivo experiments revealed that, even when properly aligned, the EAM points were often localized to positions outside the MRI-defined endocardial LV surface. There are a number of potential causes of this difference. First, actual volume differences exist between the times the MRI and the cardiac mapping were performed, both as a result of the fluid status and potential cardiac dysfunction due to the length of time the animal was under anesthesia. Second, during LV catheter mapping, it is possible to deflect the catheter such that the actual LV chamber at the point of contact is deformed. Third, although both the MRI and EAM acquisitions were gated to end-diastole, the delayed enhancement MRI pulse sequence actually acquires the images over a period of time in late-diastole—not at a single time point as with EAM.
Despite these potential errors, there are several pieces of evidence to suggest that the accuracy of the registration process may yet be clinically useful in the in vivo situation. Based upon the registered MRI images: 1) the ablation catheter could be navigated to the mitral valve annulus (a unique structure with a characteristic electrophysiologic signature); 2) radiofrequency ablation lesions could be placed within ∼2 mm from endocardial FeO injection “targets”; and 3) ablation lesions were accurately and reproducibly placed along the borders of the chronic myocardial scar. This is particularly significant, because only the registered MR image was employed to guide the ablation catheter during these experiments. It is important to note that during an actual clinical situation, the registered imaging data are unlikely to be used alone, but rather in concert with the contact electrophysiologic and electroanatomic information. These further refinements in catheter localization are expected to further increase mapping accuracy beyond that noted in these experiments.
A number of MRI pulse sequences were used in this study. To visualize the aorta, a 3D contrast-enhanced MR angiogram is a rapidly acquired pulse sequence that sharply defines the endoluminal boundary of the vessel. The LV endocardial boundary was defined using short- and long-axis cine acquisitions that provide sharp contrast between the blood pool and myocardium. Use of the end-diastole frames of the cine sequences allows the best approximation to the timing of the point acquisition during EAM. The delayed enhancement pulse sequence was also employed in the porcine infarction experiments to visualize the scarred tissue. All of these pulse sequences could be readily performed in a clinical cardiac evaluation protocol.
It is not uncommon to have unusual ventricular anatomy owing to an aneurysm, and catheter manipulation within the cardiac chamber would be greatly facilitated with a properly registered high-fidelity MRI dataset. However, the greatest potential utility of image integration is to have a direct understanding of the location and extent of the scarred myocardial substrate. The contact electrograms could then be employed to further refine catheter manipulation to critical portions of the VT circuits within the scar. The registered image would be most useful for substrate mapping during sinus rhythm. In the presence of tachyarrhythmias, the chamber geometry could be different, thus obviating the utility of MRI image integration.
In this report, the MR images were manually segmented to delimit the aorta, LV endocardium, and scar. Manual segmentation is a time- and labor-intensive process that somewhat limits the practical utility of this approach. However, a number of automated segmentation algorithms do exist and could be optimized for use with these datasets. Also, because of the finite number of MRI slices obtained per ventricle in each experiment, the final segmented reconstruction has a somewhat “choppy” artificial appearance, which could be ameliorated in part by employing a surface “smoothing” algorithm.
The presence of any volume discrepancy between the 3D MRI and EAM datasets noted in the in vivo experiments must be studied in clinical cases to determine its clinical significance. Furthermore, alternative rhythms such as tachyarrhythmias or pacing may alter the LV geometry, thus potentially obviating the utility of the registered 3D MRI dataset. Similarly, any biologic or non-biologic “noise” that degrades the quality of the electroanatomic map (e.g., location inaccuracies resulting from respiratory motion) may impact negatively on the final integrated image.
In patients with severe atherosclerotic aortic disease, extensive catheter manipulation within the aorta is undesirable because of the risks of peripheral embolization. However, further work may reveal that mapping only a portion of the aorta may be sufficient to allow for proper registration, thereby minimizing the amount of catheter manipulation performed in the aorta.
With current technology, MRI cannot be performed in patients with pre-existing pacemakers or implantable defibrillators. Because patients with VT are being increasingly treated with these devices, the use of this image integration strategy in these patients may be limited. However, pacemaker and implantable cardioverter-defibrillator lead technology is rapidly evolving, and it is conceivable that MRI-compatible leads may be developed. Another potential option is to perform CT scanning. Although it does not allow one to directly visualize the infarcted tissue, CT scanning can provide precise chamber geometry detail, including thinning of the myocardium in regions of transmural infarction. This was not specifically examined in this study, but it should be feasible to register 3D CT imaging datasets with real-time EAM.
These in vitro simulation and in vivo porcine experiments establish the proof-of-principle that pre-acquired cardiac MR images can be properly registered with intra-procedural EAM data. This work represents the first reported experimental evidence of the feasibility of guiding catheter ablation of the LV using MRI. Further work is required to determine the clinical utility of this image integration strategy in guiding catheter-based substrate mapping and ablation of VT.
The authors thank Dr. Petr Neuzil for his critical review of this manuscript.
This work was supported in part by Biosense-Webster, Inc. and by an NIH K23 award (HL68064-02) to Dr. Reddy.
- Abbreviations and acronyms
- computed tomography
- electroanatomic mapping
- field of view
- iterative closest points
- left ventricle/ventricular
- magnetic electroanatomical mapping
- myocardial infarction
- magnetic resonance
- magnetic resonance imaging
- number of excitations
- slice width
- echo time
- repetition time
- ventricular tachycardia
- Received February 28, 2004.
- Revision received August 14, 2004.
- Accepted August 23, 2004.
- American College of Cardiology Foundation
- Marchlinski F.E.,
- Callans D.J.,
- Gottlieb C.D.,
- et al.
- Soejima K.,
- Suzuki M.,
- Maisel W.H.,
- et al.
- Reddy V.Y.,
- Neuzil P.,
- Taborsky M.,
- et al.
- Arenal A.,
- Glez-Torrecilla E.,
- Ortiz M.,
- et al.
- Callans D.J.,
- Ren J.F.,
- Michele J.,
- et al.
- Reddy V.Y.,
- Wrobleski D.,
- Houghtaling C.,
- et al.
- Kim R.J.,
- Fieno D.S.,
- Parrish T.B.,
- et al.
- Gepstein L.,
- Hayam G.,
- Ben-Haim S.A.
- Vale P.R.,
- Losordo D.W.,
- Milliken C.E.,
- et al.
- ↵Schroder WJ, Martin KM, Lorensen WE. The Visualization Toolkit: An Object-Oriented Approach to 3D Graphics. 2nd edition. Prentice-Hall, 1997.