Image integration using NavX fusion: Initial experience and validation

Image integration using NavX fusion: Initial experience and validation Anthony G. Brooks, PhD, Lauren Wilson, BSc, Pawel Kuklik, MSc, Martin K. Stiles, MBChB, Bobby John, MD, Shashidhar, MD, Hany Dimitri, MBBS, Dennis H. Lau, MBBS, Ross L. Roberts-Thomson, BMedSc, Christopher X. Wong, Glenn D. Young, MBBS, Prashanthan Sanders, MBBS, PhD From the Cardiovascular Research Centre, Department of Cardiology, Royal Adelaide Hospital and the Disciplines of Medicine and Physiology, University of Adelaide, Adelaide, South Australia. BACKGROUND Three-dimensional virtual anatomic navigation is increasingly used during mapping and ablation of complex arrhythmias. NavX Fusion software aims to mold the virtual anatomy to the patient’s computed tomography (CT) image; however, the accuracy and clinical usefulness of this system have not been reported. OBJECTIVE The purpose of this study was to assess the accuracy and describe the initial experience of CT image integration using NavX Fusion for atrial fibrillation ablation. METHODS This study consisted of 55 consecutive patients undergoing atrial fibrillation ablation using NavX Fusion navigation. Left atrial NavX geometries were compared to a corresponding CT for geometric match. Geometric match, expressed as the difference in millimeters between CT and NavX geometry, was calculated for the original geometry (GEO-1), field scaled and primary fused geometry (GEO-2), and final secondary fused geometry (GEO-3). Navigational accuracy was assessed by moving the catheter to 10 discrete anatomic sites and determining the distance between the

Introduction Catheter ablation is increasingly used as a treatment modality in patients with drug-refractory atrial fibrillation (AF). Most ablation strategies have a central theme of pulmonary vein isolation. However, additional substrate modification is Dr. Brooks is supported by the Mary Overton Award from the Royal Adelaide Hospital. Dr. Stiles is supported by the National Heart Foundation of New Zealand and the Dawes Scholarship, Royal Adelaide Hospital. Dr. John is supported by the Biosense Webster Electrophysiology Scholarship, University of Adelaide. Dr. Dimitri is supported by the Cardiac Society of Australia and New Zealand. Dr. Lau is supported by the Earl Bakken Electrophysiology Scholarship, University of Adelaide, and a Kidney Health Australia Biomedical Research Scholarship. Dr. Sanders is supported by the National Heart Foundation of Australia. Dr. Sanders reports having served on the advisory board of, and having received lecture fees and research funding from, St. Jude Medical, Bard Electrophysiology, and Biosense Webster. Address reprint requests and correspondence: Dr. Prashanthan Sanders, Cardiovascular Research Centre, Department of Cardiology, Level 5, McEwin Building, Royal Adelaide Hospital, Adelaide, SA 5000, Australia. E-mail address: prash.sanders@adelaide. edu.au. (Received December 10, 2007; accepted January 4, 2008.)

catheter tip and the closest GEO-2, GEO-3, and CT surface. Fusion integration time and procedural and fluoroscopic durations were recorded to assess clinical usefulness. RESULTS GEO-1, GEO-2 and GEO-3 were associated with CT–GEO errors of 6.6 ⫾ 2.8 mm, 4.1 ⫾ 0.7 mm, 1.9 ⫾ 0.4 mm, respectively. Navigational accuracy was not significantly different for GEO-2, GEO-3, and CT at 3.4 ⫾ 1.6 mm to any surface. A significant (P ⱕ.001) inverse curvilinear relationship was present between case number and the time required for image integration (r2 ⫽ 0.35) and the fluoroscopic time normalized for procedural duration (r2 ⫽ 0.18). CONCLUSION Image integration using the NavX Fusion software is highly accurate and is associated with a progressive reduction in fluoroscopic time relative to procedural duration. KEYWORDS Three-dimensional image integration; Computed tomography; Atrial fibrillation (Heart Rhythm 2008;5:526 –535) © 2008 Heart Rhythm Society. All rights reserved.

required to improve outcomes in some patients with paroxysmal AF and in almost all patients with persistent or permanent AF. Therefore, accurate visualization of the left atrial anatomy and its variants is increasingly important in AF ablation. Anatomic variants with possible implications to AF ablation include left atrial pouches, ridges, and differences in vein number, departure angle, and ostium locations.1– 8 These more subtle anatomic deviations are poorly defined by conventional imaging and may contribute to prolongation of radiofrequency, procedural, and fluoroscopic times. Three-dimensional (3D) virtual anatomic navigation has progressively evolved to assist complex AF ablation. Reconstructed 3D geometry using electroanatomic mapping has been demonstrated to be associated with significantly reduced fluoroscopic and procedural times for AF ablation.9 –11 To improve visualization of the highly variable left atrial anatomy, significant efforts have been made to integrate electroanatomic mapping with computed tomographic (CT) or magnetic resonance imaging (MRI) surface geom-

1547-5271/$ -see front matter © 2008 Heart Rhythm Society. All rights reserved.

doi:10.1016/j.hrthm.2008.01.008

Brooks et al

NavX Fusion Image Integration

etry. Navigation using the CT image may assist the operator in appreciating subtle left atrial anatomic variants that often are not apparent on electroanatomic mapping. Until now, image integration has been able to “merge” electroanatomic mapping by superimposing the electroanatomic map onto a reconstructed CT and hence locate in real time the ablation catheter within this anatomically correct structure. Early experience using this system in most series demonstrated a degree of accuracy within the range required for clinical use and with potentially important outcomes.12–16 An advance in image integration entitled “NavX Fusion” (St. Jude Medical, St. Paul, MN, USA) recently has become available. Unlike previously available technology, the process of fusion dynamically molds the created geometry to the CT. Although conceptually this may be considered a useful strategy, several potential confounding factors could limit the accuracy of such a process, mandating appropriate validation of its accuracy prior to widespread clinical use. In this report, we describe the first clinical experience and validation of image integration using the NavX Fusion software.

Methods Study population The study consisted of 55 consecutive patients with symptomatic-drug refractory AF undergoing ablation guided by NavX Fusion. Baseline characteristics of the patients are given in Table 1. The study protocol was approved by the institutional Clinical Research and Ethics Committee, and all patients provided written informed consent for the procedure. All antiarrhythmic drugs, with the exception of amiodarone, were ceased at least five half-lives before the study. Prior to the procedure, all patients received anticoagulation with warfarin (international normalized ratio 2– 4) for at least 6 weeks and underwent transesophageal echocardiography to exclude left atrial thrombus.

Cardiac CT imaging Prior to the ablation procedure, all patients underwent cardiac imaging with a Siemens Sensation 64 slice spiral CT Table 1

Baseline characteristics of all patients (n ⫽ 55)

Age (years) Male [n (%)] Body mass index (kg/m2) Longest AF episode㛳 Duration of AF (months)㛳 Type of AF Paroxysmal Persistent Permanent No. of antiarrhythmic drugs failed Structural heart disease Left atrial parasternal size (mm) Left atrial area (cm2) LV end-diastolic diameter (mm) LV end-systolic diameter (mm) LV ejection fraction (%) 㛳

56 ⫾ 11 37 (67%) 29 ⫾ 4 48 days (12 hours–12 years) 48 (7–360) 29 (53%) 14 (25%) 12 (22%) 1.8 ⫾ 0.9 14% 4.3 ⫾ 0.7 24 ⫾ 7 50 ⫾ 7 32 ⫾ 8 59 ⫾ 10

AF ⫽ atrial fibrillation; LV ⫽ left ventricular. [median (range)].

527 scanner (Siemens AG, Munich Germany) prior to the ablation procedure. The scanner has 32 channels with 0.6-mm collimation and oversampling in the Z-axis with flying focal spot to yield an effective 64 detector rows. Gantry rotation time was 370 ms with an x-ray tube potential of 120 kV (pitch 0.24). A dose modulation algorithm was used to optimize tube currents at approximately 500 mA. The scan was conducted in a caudocranial direction during a mild inspiration breathhold. A 70-mL bolus of Isovue 370 was introduced into the antecubital vein at a rate of 5 mL/s, followed by 50 mL of saline. The scanner featured “bolus tracking,” which enabled left atrial images to be obtained during optimal opacification. Images were gated to a mean 61 ⫾ 15% of peak QRS for patients in sinus rhythm and arbitrarily gated to 45% of peak QRS for patients in AF. The left atrium was segmented and reconstructed in three dimensions from CT slices using EnSite Verismo software (St. Jude Medical).

Electrophysiologic study Electrophysiologic study was performed in the postabsorptive state with midazolam and fentanyl sedation. The following catheters were utilized for mapping: (1) 10-pole catheter (2-5-2 mm interelectrode spacing, Daig Electrophysiology, St. Jude Medical) positioned within the coronary sinus with the proximal bipole at the coronary sinus ostium as determined in the best septal left anterior oblique position, (2) 10-pole circumferential mapping catheter (Lasso, Biosense Webster, Diamond Bar, CA, USA), and (3) 3.5-mm-tip externally irrigated ablation catheter (Celsius ThermoCool, Biosense Webster). The left atrium was accessed using a single transseptal puncture, after which a bolus of unfractionated heparin (100 units/kg) was administered. Repeat boluses were infused periodically to maintain the activated clotting time between 250 and 350 seconds.

Three-dimensional geometry creation The 3D geometry of the left atrium was created using NavX Fusion. In brief, NavX navigation is an impedance-based measure derived from the voltage gradient that appears across the tissue when a current is applied through a pair of surface electrodes. For 3D navigation, six surface electrodes are placed in three pairs: anterior to posterior, left to right lateral, and superior (neck) to inferior (left leg). The three electrode pairs form three orthogonal axes (X-Y-Z), with the heart at the center. A “known” 5-kHz current is alternately delivered through each pair of surface electrodes to form a 3D transthoracic electrical field. The absolute range of voltage along each axis varies from each other, depending on the volume and type of tissue subtended between each surface– electrode pair. The voltage gradient is divided by the known applied current to determine the impedance field that has equal unit magnitudes in all three axes. Each level of impedance along each axis corresponds to a specific anatomic location within the thorax. As catheters are maneuvered within the chambers, each catheter electrode

528

Heart Rhythm, Vol 5, No 4, April 2008

Figure 1 Screenshot series showing progressive improvement in geometry similarity to computed tomography for original geometry (GEO-1), field scaled and primary fused geometry (GEO-2), and secondary fused geometry (GEO-3). Note significant dragging together of the surfaces during the secondary fusion stage.

senses the corresponding levels of impedance, derived from the measured voltage. Timed with the current delivery, NavX calculates the X-Y-Z impedance coordinates at each catheter electrode and expresses them in millimeters to graphically locate the catheters in real time to enable nonfluoroscopic navigation. For this study, moderate catheter responsiveness was selected to filter high-frequency artifact introduced by atrial systole, diastole, or fibrillation. Respiratory compensation was collected just before mapping to filter low-frequency cardiac shift associated with the breathing cycle. A proximal coronary sinus electrode was selected as a positional reference to which all geometries and catheter positions were referenced. The geometry was collected continuously and not gated to ECG or respiratory phase so that the final geometry integrated the filtering effects of catheter sensitivity and respiration adjustment. The Lasso catheter was used to “outline” gross atrial anatomy (body and veins). The ablation catheter was used to define pulmonary vein branches and pulmonary vein ostia and to add definition to the left atrial body. The CT reconstruction, visualized on a split screen, was used to guide finer anatomic definition with the ablation catheter. Upon completion, maps were edited to eliminate “false space” (i.e., geometry with sparse geometry points) and erroneous structure definition. An example of the original geometry (GEO-1) is shown in Figure 1.

Correction for impedance nonlinearity (field scaling) After the geometry was edited, “field scaling” was applied to the geometry. Field scaling compensates for variations in impedance between the heart chambers and venous structures. In broad terms, the electrical attenuations of tissues surrounding the heart chambers differ from each other and from the attenuation within the blood-filled chambers. Differences among the structures surrounding the chambers are significantly greater than those within the chambers. Such regional differences in attenuation create a corresponding nonlinearity in the impedance field across the thorax. Before field scaling is applied, a linear impedance field is assumed that results in a distortion of the X-Y-Z coordinates when a “roving” catheter is maneuvered among the differing regions of impedance. Accordingly, as the catheter is maneuvered, the displayed image of the catheter seems stretched within the pulmonary veins while appearing more normally proportioned in the center of the left atrium. Using the “actual” electrode spacing constant of the catheter acquiring the geometry, data on the degree of departure from linearity are collected at each geometry point in 3D space. When applied, the field scaling algorithm adjusts the 3D coordinates for nonlinearity with respect to the measured catheter spacing. The final result is a geometry and navigational space that is more physiologically relevant and more closely resembles the CT (Figure 1, Field scaling).

Brooks et al

NavX Fusion Image Integration

529

Figure 2 Screenshot series showing the effect of secondary fusion on geometry and navigation space in a redo atrial fibrillation ablation patient with a remodeled left atrium from a previous procedure. Secondary fusion enabled effective navigation within the original computed tomogram scanned 415 days prior to the redo procedure. Panels from left to right indicate GE01 field scaled and primary fused geometry (GEO2), secondary fusion process demonstrating fiducial points, and secondary fused geometry (GEO3).

Image integration (fusion) The field scaled geometry is “fused” to the CT in two stages, termed primary fusion and secondary fusion (Figure 1). Primary fusion uses three fiducial (i.e., landmark) corresponding points on the created geometry and the CT image to superimpose, or lock together, both structures (Figure 1, GEO-2). These points are chosen to ensure reasonable 3D anatomic separation. Based on previous experience with the CARTO merge software (Biosense Webster), anterior left atrial points were avoided for initial image registration due to the inherent mobility of this structure. Following primary fusion, a process of secondary fusion was performed (Figure 1, GEO-3). Secondary fusion points or “fiducials” were applied to the primary fused geometry (GEO-2) at sites of local mismatch between the two superimposed geometries. In this unique component of image

fusion, the created geometry surface is molded to the CT surface while also “bending” the 3D navigation space within the geometry. For example, if the anterior wall of the geometry is fused anteriorly to the CT, the catheter location also will be moved so that, when the catheter is repositioned on the anterior wall post fusion, it should be visualized on the CT surface and not within the bounds of the original prefusion geometry (Figure 2). Upon completion of primary and secondary fusion, the geometry display was disabled, and all lesions, labels, and catheters were projected on the CT image (Figure 3).

Validation of NavX fusion Validation of geometric match and catheter navigational accuracy was assessed in the first 15 patients of the study. CT and procedural details are given in Table 2.

Figure 3 Screenshot of endoscopic and whole atrium views used for pulmonary vein isolation using NavX Fusion. Location of the catheters during ablation of the left common pulmonary vein are shown.

530 Table 2

Heart Rhythm, Vol 5, No 4, April 2008 Validation patient procedural details (n ⫽ 15)

CT Details Time prior to procedure (days)㛳 Rhythm at scan SR AF Procedural Details Time (minutes) required for map creation* Time (minutes) required for fusion† No. of veins isolated Pulmonary vein isolation Roof line Mitral isthmus line Fractionated electrogram ablation Tricuspid isthmus line Total procedural time (min)‡ Validation Details Rhythm at fusion SR AF Different rhythm at fusion compared with CT? Yes No Time between fusion and validation (min)

25 (1–645) 69% 31% 22.1 ⫾ 12.7 ⫾ 3.7 ⫾ 100% 73% 20% 40% 13% 192 ⫾

11.3 2.6 0.5

The CT–GEO distance for the entire left atrium for each of three geometry conditions (GEO-1, GEO-2 and GEO-3) was calculated. Mean, SD, and range consequently were used to describe the CT–GEO distance. In addition, CT– GEO distance for each pulmonary vein, left atrial appendage, and anterior, posterior, septal, lateral, and roof component of the left atrial body was calculated. The raw CT– GEO distance data (n ⫽ 98,529) are presented in histograms for each geometry condition in Figure 4.

Navigational validation

60

33% 67% 40% 60% 140 ⫾ 32

AF ⫽ atrial fibrillation; CT ⫽ computed tomography; SR ⫽ sinus rhythm. *Including geometry creation and editing. †Fusion requires stepwise addition of fiducial points to merge CT and geometry together. ‡Transseptal puncture to completion of ablation. 㛳 [median (range)].

The CT to GEO (CE–GEO) distance was calculated at each stage of NavX geometry (GEO-1, GEO-2, or GEO-3). These data were used to estimate the relative change in geometry with respect to the “gold standard” structure of the CT. The accuracy of navigation within GEO-2, GEO-3, or the CT was assessed at 10 anatomically distinct sites via live estimation of the catheter tip to surface distance.

In order to assess the accuracy of navigation using the geometries created, the distance between the catheter tip and the closest GEO-2, GEO-3, or CT surface was recorded in real time using the NavX system. The ablation catheter was sequentially navigated to the following sites: each pulmonary vein ostium, mid–left atrial roof, midposterior left atrium, anterior mitral annulus, left atrial fossa, lateral mitral isthmus, and anterior base of the left atrial appendage. Once the system indicated a stable catheter position, a combination of fluoroscopic movement and electrogram verification of endocardial wall contact was established and the distance to the surface recorded. While the catheter was held steady, the three geometries (GEO-2, GEO-3, and CT) were cycled through, and the distance from catheter tip to selected surface was recorded. In order to determine the adequacy of the system throughout a complex procedure encompassing all potential sources of error, navigational validation was conducted at the completion of the clinical procedure.

Clinical usefulness of NavX fusion

Image integration assessment (CT to GEO distance)

Fusion time and fluoroscopic and total procedural durations were recorded for all patients. The number of primary and secondary fiducials applied to the field scaled geometry was also recorded for all patients. The time required for fusion of the geometry to the CT was recorded from completion of

The distance between the CT and the (1) non–field scaled (GEO-1), (2) field scaled and primary fused (GEO-2), and (3) secondary fused (GEO-3) geometries was determined using custom-designed and previously validated software with the ability to perform geometric analysis of 3D objects (Medicalgorithmics, Austin, TX USA).17 This software reproduced the CT and geometry in the same 3D space and orientation as presented in NavX. The CT was segmented to the primary branching of the pulmonary veins, and the mitral valve orifice was “cut” from the CT surface. The left atrium match statistics were calculated for the whole left atrium and for the pulmonary veins, left atrial appendage, and anterior, posterior, septal, lateral, and roof left atrial body segments. The number of CT triangles per left atrium was 6,569 ⫾ 1,441, with an associated mean triangle area of 2.6 ⫾ 1.3 mm2. Euclidean distance (linear distance between two points in 3D space) between a given CT triangle and the closest NavX geometry triangle was calculated for each CT triangle and summarized using conventional methods.

Figure 4 Raw data histograms for CT–GEO distance for three geometry conditions. Lines and numbers indicate median value.

Brooks et al

NavX Fusion Image Integration

GEO-1 to completion of field scaling and secondary fusion, whereby catheters could be visualized and navigated within the CT reconstruction. Given the variation in the extent of ablation used in individual cases (detailed later), fluoroscopic time relative to total procedural time (transseptal to completion of ablation) was used as an indicator of fluoroscopic dependence for AF ablation.

Ablation procedure Ablation was performed using a combination of fluoroscopic and NavX navigation. Both the whole and endoscopic views of the CT geometry were used to aid navigation and ablation (Figure 3). All patients underwent circumferential pulmonary vein ablation with an endpoint of isolation confirmed by circumferential mapping with either elimination or dissociation of pulmonary venous potentials. Pulmonary vein ablation was commenced randomly in either the right or left vein and performed individually or as a pair when the ostia were coalescent. Ablation of the pulmonary veins was performed using a delivered power of 30 W with irrigation rates of 30 mL/min. Additional substrate modification by linear ablation (roof line or mitral isthmus ablation) or targeting regions of complex fractionated atrial electrograms was performed in patients with AF episodes persisting for more than 48 hours, structural heart disease, or marked left atrial dilation (longitudinal diameter ⬎57 mm). Cavotricuspid isthmus ablation with an endpoint of bidirectional isthmus block was performed only in patients with a history of typical flutter or if mapping confirmed typical flutter during the procedure. The endpoint of substrate modification was either electrophysiologically confirmed linear conduction block established via pacing maneuvers or elimination of fractionation. Substrate modification was performed using a delivered power of 30 to 35 W with irrigation rates of 30 to 60 mL/min.

Statistical analysis Normally distributed data are expressed as mean ⫾ SD. Significantly skewed data (P ⬍.05, Kolmogorov-Smirnov test) are summarized as median and interquartile range. Categorical data are expressed as count and percentages. CT–GEO distance data were analyzed using a linear mixed effects model with anatomic region (anatomic region 1–10), stage of geometry (GEO-1, GEO-2, GEO-3), and their interaction (GEO*region) modeled as fixed effects. Random effects were added to the model to account for nested data (i.e., three fusion stages and 10 regions within each patient). Post hoc tests revealed sources of significance in the mixed effects model. If all data were considered independent, there would be an increased probability of obtaining false significance. Data from navigational validation were analyzed in the same fashion, with anatomic region (anatomic region 1–10) and reference geometry (GEO-2, GEO-3, and CT) modeled within a linear mixed effects framework.

531 The learning curve and clinical usefulness were assessed by determining the relationship between case number (i.e., relative experience with NavX Fusion) and (1) fluoroscopic time, (2) procedural time, (3) relative fluoroscopic dependence, (4) fusion time, and (5) number of fiducials explored via scatter plots for the whole group and unpaired t-tests when comparing the first and last 15 patients in the study. P ⬍.05 was considered significant.

Results Three primary fiducial points were used to register the field scaled geometry with the CT image. Primary fiducial points were located at clearly defined pulmonary vein ostia (67%), superior base of the left atrial appendage (13%), roof (13%), and posterior wall (7%). Due to the relatively high definition of the field scaled geometry, no further imaging or angiography was used to define the accuracy of these primary fusion points. In cases of local mismatch between the two geometries, points of secondary fusion at a mean 44 ⫾ 19 points (range 15– 81 points) were used to dynamically register the geometry to the CT. CT scan was performed a median 25 days (range 1– 645 days) before the ablation procedure. Nine patients who were more than 60 days between CT scan and AF ablation were undergoing repeat procedures; therefore, repeat CT scans were not performed to minimize additional exposure to radiation.

Image integration assessment (CT–GEO distance) A mean of 6,569 ⫾ 1,441 CT–GEO error measurements were recorded for each left atrial and geometry condition. The raw (n ⫽ 98,529) mean distance for CT to GEO-3 was 2.6 ⫾ 2.2 mm (median 1.8 mm, interquartile range 1.1–3.3 mm; Figure 4). To summarize the highly skewed raw data (Kolmogorov-Smirnov test: P ⬍.001; Figure 4), the median CT–GEO distance was calculated for each patient’s overall and regionalized left atria. Using the CT geometry as a gold standard reference, the accuracy of the original NavX geometry (GEO-1) improved significantly with field scaling and primary fusion (GEO-2) and again with secondary fusion (GEO-3). The distance (n ⫽ 15) between the geometry and CT reduced significantly (P ⬍.001) from 6.6 ⫾ 2.8 mm (median [interquartile range] ⫽ 6.1 [4.0 –9.1]; GEO-1) to 4.1 ⫾ 0.7 mm (median [interquartile range] ⫽ 4.1 [3.6 – 4.6]) with the application of field scaling and primary fusion (GEO-2) and to 1.9 ⫾ 0.4 mm (median [interquartile range] ⫽ 1.7 [1.6 –2.4]) with secondary fusion (Figure 5A, GEO-3). The CT–GEO distance was significantly different for the 10 anatomic regions (P ⬍.001), with the lowest error in the left atrial roof and pulmonary veins (P ⬍.001) and the larger errors in the left atrial septal wall, lateral wall, and appendage (P ⬍.03; Figure 5A). A nonsignificant interaction (stage of geometry and anatomic location; P ⫽ .1) suggested that the initial GEO-1 regional errors were attenuated similarly

532

Heart Rhythm, Vol 5, No 4, April 2008

Figure 5 A: Registration accuracy. CT–GEO difference for each region across three geometry conditions. B: Navigational accuracy. Distance between catheter tip and surface of GEO-2, GEO-3, and CT. ANT ⫽ anterior; LAA ⫽ left atrial appendage; LAT ⫽ lateral; LIPV ⫽ left inferior pulmonary vein; LSPV ⫽ left superior pulmonary vein; POST ⫽ posterior; RIPV ⫽ right inferior pulmonary vein; ROOF ⫽ roof; RSPV right superior pulmonary vein; SEPT ⫽ septal.

with field scaling/primary fusion (GEO-2) and secondary fusion (GEO-3). Forty percent of patients had a different rhythm at CT scan compared with map development and fusion. These patients did not have significantly different CT–GEO distance than those who had the same rhythm for both (P ⫽ .85).

Secondary fusion with NavX The mean number of secondary fusion points applied to the anterior, posterior, lateral, septal, and roof of the left atrial body were 6.5 ⫾ 3.7, 5.7 ⫾ 3.0, 2.2 ⫾ 2.2, 2.5 ⫾ 2.3, 2.0 ⫾ 1.8, respectively. In comparison, a mean 6.4 ⫾ 3.9 points was used to fuse each of the pulmonary veins or left atrial appendage. One-way repeated measures analysis of variance demonstrated significant regional difference in the number of secondary fusion points (P ⬍.001), with more fusion points applied to the pulmonary veins and anterior left atrium in our initial experience. A weak inverse association was noted between the number of secondary fiducial points applied to the field scaled geometry (conversion of GEO-2 to GEO-3) and the percent change in error to the reference CT geometry (r ⫽ – 0.21, P ⫽ .01).

Navigational validation When stabilized at 10 anatomically distinct locations within the left atria, the mean distance from the ablation catheter tip to the GEO-2 (3.6 ⫾ 1.5 mm), GEO-3 (3.3 ⫾ 1.4 mm),

or CT surface (3.4 ⫾ 1.6 mm) did not change significantly (P ⫽ .5; Figure 5B). The mixed linear model revealed no significant regional differences in catheter to surface error (P ⫽ .2) or in the interaction term between geometry stage and region (P ⫽ .9). Patients with a different rhythm at CT scan compared with map development and fusion did not have significantly different catheter to surface error than those who had the same rhythm for both (P ⫽ .26).

Learning curve and clinical assessment of NavX fusion A significant inverse curvilinear relationship was observed between case number and time required for fusion (r2 ⫽ 0.35, P ⬍.001; Figure 6A), such that fusion time was significantly longer for the first 15 patients (12.7 ⫾ 2.4 min) than for the last 15 patients (8.6 ⫾ 2.6 min; P ⬍.001). However, there was no significant reduction in the number of primary and secondary fiducials used for fusion over this experience (P ⫽ .17). A significant inverse curvilinear relationship was observed between NavX Fusion case number and the ratio of fluoroscopic time to total procedural time (r2 ⫽ 0.18, P ⫽ .001; Figure 6B), such that there was a significant reduction in the mean fluoroscopic time relative to total procedural duration for the last 15 patients (0.34 ⫾ 0.05) compared with the first 15 patients (0.43 ⫾ 0.11, P ⫽ .009). However,

Brooks et al

NavX Fusion Image Integration

533 ●

Greater NavX Fusion experience was associated with a significant decline in the time required for CT integration and fluoroscopic dependence.

This latest addition to the repertoire of navigational tools available to the electrophysiologist is highly accurate and facilitates nonfluoroscopic navigation within an anatomically accurate virtual structure.

Efficacy and validation of virtual cardiac navigation in AF ablation

Figure 6 Scatter plots showing decline in time required for CT integration (A) and reduction in fluoroscopic dependence relative to procedural duration (B).

there was no significant relationship between NavX Fusion case number and fluoroscopic (P ⫽ .37) or procedural duration (P ⫽ .14).

Discussion This report describes the validation and clinical usefulness of the new image integration software NavX Fusion, which fuses or molds the virtual anatomy to the CT surface. The study demonstrates the following: ●

●

Field scaling and primary and secondary fusion significantly improve approximation of the NavX geometry to the CT. Each step resulted in progressive improvement in the association between the created geometry and the CT. The final structure was associated with a mean error of 1.9 ⫾ 0.4 mm to the CT (range 1.3–2.6 mm), well within the range that would be considered clinically relevant. The change in geometry associated with secondary fusion can be used to accurately localize catheters within the geometry. This is because, in addition to morphing the shell of the structure, fusion effectively applies the same algorithm to the 3D navigation space (Figure 2). Independent of GEO stage, the accuracy of navigation within NavX geometry is maintained. Mean absolute catheter error over 10 anatomically discrete locations of the left atrium was 3.4 ⫾ 1.6 mm.

Although early clinical validation studies on 3D mapping systems (CARTO or NavX) have suggested a limited impact on pulmonary vein isolation success rates,9 –11 the evolution of more complex AF ablation that uses anatomic lines and focal burns at sites of fractionation or highest dominant frequency would be difficult to perform without the assistance of virtual navigation. Furthermore, there is compelling evidence for reduced fluoroscopy dependence using these systems.9 –11 These systems provide a permanent record of where ablation was performed, and they can be used to present timing and signal data, which are critical for testing anatomic lines of block, mapping flutter circuits, and directing focal complex fractionated electrogram ablation. A 3D navigation system to assist the targeted substrate ablation of persistent and permanent AF is an important component of a modern electrophysiology laboratory. To improve visualization of the highly variable atrial anatomy and pulmonary venous structure, significant efforts have been made to integrate the electroanatomic map to CT/MRI surface geometry. Albeit an appealing concept, merging virtual geometry and CT/MRI anatomy is scan and operator dependent. It relies upon the subjective selection of matching “landmark” pairs on CT and virtual geometry to bring together the two structures. In addition, differences in timing, respiratory phase, rhythm, hemodynamic status, and heart rate between the CT scan and the development of virtual geometry are potential sources of CT–GEO error,18 with resultant navigational inaccuracy. Regardless of these limitations, CARTO merge CT integration has been associated with reduced fluoroscopic times and suggestions of improved clinical success compared with CARTO alone.13,14 This study is the first to describe the further iteration of image integration using NavX Fusion, which has the unique feature of molding the geometry to the CT/MRI surface.

Registration accuracy (CT–GEO distance) Field scaling and primary and secondary fusion are new features that allow the NavX geometry to be adjusted for nonlinearity, anchored and locally fitted to a reconstructed CT. Field scaling adjusts for the nonlinear impedance field over the left atrium in reference to a known constant of catheter electrode spacing. In effect, field scaling of the left atrial geometry (GEO-2) consistently resulted in a more anatomically correct left atrium than that mapped originally (GEO-1). NavX Fusion encompasses the procedure of adding primary fiducials to initially anchor the geometry to the

534 CT and then multiple secondary fiducial points to locally fuse the geometry to the CT structure (Figure 1). Using the CT as a gold standard reference, both raw (n ⫽ 98,529) and summary (n ⫽ 15) CT–GEO distance data demonstrate a stepwise improvement in original GEO morphology (GEO-1) with (1) field scaling and primary fusion (GEO-2) and with (2) secondary fusion (GEO-3, Figure 1). Adjustment of the geometry by field scaling and primary fusion resulted in a 40% improvement in match error, whereas secondary fusion with a mean 44 ⫾ 19 fiducials further reduced the error by 54%. Given that the original geometry is equivalent to that 19 produced by Version 6 software, NavX Fusion represents a significant improvement in anatomical relevance. A change in cardiac rhythm between the CT scan and fusion was not associated with increased geometry error. Data on the final NavX geometry (GEO-3) demonstrate that on average the geometry departs from the CT by 1.9 ⫾ 0.4 mm (range 1.3–2.6 mm). Given the complexity of CT left atrial anatomy, this represents high concordance between NavX and CT geometries. These results compare favorably to data obtained with the CARTO system using surface registration, where published geometric differences range from 1.7 to 2.7 mm.13–16,19,20 Given that GEO-2 (field scaled and primary fused) is similar to the landmark-registered CARTO merge image, NavX match statistics are lower (n ⫽ 15, mean error 4.1 ⫾ 0.7 mm) compared with 6.4 mm16 and 5.6 mm20 mean errors reported for CARTO merge. Finally, CT to GEO-3 errors were approximately 1.7 mm in the veins compared with approximately 6 mm errors reported by Kistler et al16 for 39-point left atrial CARTO maps.

Navigational accuracy With real-time 3D mapping systems, the generated geometric map should resemble the CT image as closely as possible (minimized CT–GEO distance). Equally if not more important to the clinician is how accurately the fused map reflects the real-time position of the catheter in the heart during the ablation procedure. A high concordance between the geometric and CT maps is of limited relevance if it does not accurately represent the true location of the catheter within the heart. The navigational accuracy of NavX Fusion was validated at the end of the AF ablation procedure at a mean 140 ⫾ 32 minutes after fusion. The catheter was navigated sequentially to 10 anatomically discrete locations as determined by the operator using fluoroscopic and electrogram criteria. The distance to the field scaled geometry (GEO-2), secondary fused GEO (GEO-3), and the CT surfaces was measured. The error, or distance from catheter tip to surface, was measured as an absolute value so that positive and negative errors did not negate each other. Furthermore, the validation occurred at the end of the procedure. Therefore, the measured error integrated anatomic navigation error and that associated with impedance drift and patient movement over the case. The absolute distance to GEO-2, GEO-3, and CT surface error of 3 to 4 mm (Figure 5B) is within the

Heart Rhythm, Vol 5, No 4, April 2008 requirements for clinical procedures. Furthermore, catheter navigational error was independent of changes in cardiac rhythm between CT scan, fusion, and catheter validation. Dong et al12 reported that with assistance of radiographically and CT visible pericardial reference markers for landmark registration, CARTO merge was able to direct ablation points to within 1.8 ⫾ 1.0 mm mean accuracy of the true anatomic location as assessed postmortem. Unfortunately, the technique they used to register electroanatomic mapping and CT geometries with pericardial markers is not clinically applicable. In vivo accuracy of CARTO merge has been assessed with less impressive results when using intracardiac echocardiography as the gold standard anatomical reference.19,20 Zhong et al19 rehearsed intracardiac echocardiography-guided pulmonary vein isolation, blinded to the CARTO merge image developed just prior. The absolute error of ablation points to the nearest CT surface was 3 to 5 mm; however, the distance to the exact anatomic location (as estimated by intracardiac echocardiography) was 9 to 10 mm. Fahmy et al20 reported that ICE-guided landmark registration of the CARTO electroanatomic map and CT, resulted in a mean landmark error of 5.6 mm for posterior landmarks only, or 9.1 mm, when anterior landmarks were utilized. Furthermore, when the electroanatomic map was surface registered, landmark errors significantly increased, leading the authors to conclude that an “an accurate surface registration does not guarantee an accurate alignment with important anatomical structures.”20 This study highlighted the difference between anatomic (CT– electroanatomic mapping match) and navigational (catheter–surface error) accuracy. Taken together, the accuracy achieved by Dong et al12 with defined CT and radiographic markers with which to select landmark pairs and the relative inaccuracy demonstrated in the in vivo validations19,20 suggest that the majority of in vivo error is attributed to the incorrect selection of landmark pairs on the electroanatomic map and CT. Even though primary fusion of field scaled NavX geometry and CT is plagued by the same subjective inaccuracies as CARTO, the high-resolution structure of the NavX geometry may more accurately define particular landmark points by which to fuse the two structures, resulting in a smaller catheter navigational error.

Clinical assessment of NavX fusion In our early experience, NavX Fusion has proved to be intuitive to use, with only a short “run in time” or learning curve. The curvilinear relationship between time required for CT integration and experience demonstrates that “fusion time” has significantly reduced over our experience (Figure 6A). The fact that the same trend was not apparent for the number of fiducials applied to GEO-2 suggests that the shortened time for CT integration is due more to a faster fusion process rather than less fusion points being applied. It is apparent from our clinical experience that the major effectors of structural change, after field scaling and primary fusion, are the initial secondary fusion points. It is important to anchor the most important regions of interest, such as the

Brooks et al

NavX Fusion Image Integration

pulmonary veins, to ensure their structural alignment. Additional secondary fusion points then can be placed in the left atrial body to mold the geometry and navigation space to the CT surface. Finally, further fiducial points are used to match mapped pulmonary vein branching to that visualized on CT. The final result is catheter navigation within a CT such that navigation into individual pulmonary vein branching is possible. The inverse regional correlation between number of fiducials and change in geometry morphology (r ⫽ – 0.21, P ⫽ .01) is an indicator that the law of diminishing returns may apply to fiducial placement, such that initial fiducials elicit gross geometric structure change and consequent applications are associated with fine structural tuning. The reduced fluoroscopic dependence, as estimated by fluoroscopy to procedure time ratio, was suggestive of greater confidence in nonfluoroscopic navigation with NavX Fusion as experience with the system grew (Figure 6B); an important indicator for an increase in efficacy of NavX fusion as a nonfluoroscopic navigation system. No relationship between experience (i.e., case number) and fluoroscopic or procedural duration was observed and probably is representative of the variant procedure complexity given the tailored AF ablation approach used.

Pitfalls when using NavX Fusion Two patients were excluded from the validation cohort because applying field scaling to GEO-1 resulted in a nonphysiologic structure. The cause of the problem was that the electrode spacing of the catheter used to collect the geometry were entered incorrectly into the NavX system. Hence, the adjustment of impedance nonlinearity was made against an incorrect standard. In our experience, insufficiently mapped anatomic regions should not be fused to the CT surface because if the catheter does reach the “true endocardium,” the catheter will be projected outside the CT. Nevertheless, real-time fiducial placement at the active catheter tip can expeditiously correct this problem by “dragging” the catheter back to the CT surface. No data were excluded because of this error.

Study limitations This study performed a validation of CT–GEO closeness of fit and catheter navigational accuracy but did not evaluate the potential impact on clinical outcome. Navigational accuracy was not evaluated against a real-time imaging modality such as intracardiac echocardiography. In addition, it is important to note that CARTO match statistics form the benchmark for comparing CT–GEO errors in this study because, until now, the former has been the only system able to merge electroanatomic mapping and CT structures. CARTO and NavX CT–GEO errors are estimated using differing methodologies, so our indirect comparisons between the two should be interpreted with caution.

535

Conclusion Image integration using NavX Fusion software to mold the geometry to the CT image is highly accurate, with a mean 1.9 ⫾ 0.4 mm difference between the final geometry and the CT and a navigational error of 3.4 ⫾ 1.6 mm. Our clinical experience demonstrates reasonable CT integration times (8.6 ⫾ 2.6 min) and a reduced dependence on fluoroscopic navigation for AF ablation.

References 1. Malchano ZJ, Neuzil P, Cury RC, et al. Integration of cardiac CT/MR imaging with three-dimensional electroanatomical mapping to guide catheter manipulation in the left atrium: implications for catheter ablation of atrial fibrillation. J Cardiovasc Electrophysiol 2006;17:1221–1229. 2. Wongcharoen W, Tsao HM, Wu MH, et al. Morphologic characteristics of the left atrial appendage, roof, and septum: implications for the ablation of atrial fibrillation. J Cardiovasc Electrophysiol 2006;17:951–956. 3. Wittkampf FH, Vonken EJ, Derksen R, et al. Pulmonary vein ostium geometry: analysis by magnetic resonance angiography. Circulation 2003;107:21–23. 4. Mansour M, Holmvang G, Sosnovik D, et al. Assessment of pulmonary vein anatomic variability by magnetic resonance imaging: implications for catheter ablation techniques for atrial fibrillation. J Cardiovasc Electrophysiol 2004;15: 387–393. 5. Noseworthy PA, Malchano ZJ, Ahmed J, et al: The impact of respiration on left atrial and pulmonary venous anatomy: implications for image-guided intervention. Heart Rhythm 2005;2:1173–1178. 6. Sanders P, Stiles MK, Young GD. Virtual anatomy for atrial fibrillation ablation. J Cardiovasc Electrophysiol 2006;17:349 –351. 7. Kato R, Lickfett L, Meininger G, et al. Pulmonary vein anatomy in patients undergoing catheter ablation of atrial fibrillation: lessons learned by use of magnetic resonance imaging. Circulation 2003;107:2004 –2010. 8. Lemola K, Sneider M, Desjardins B, et al. Computed tomographic analysis of the anatomy of the left atrium and the esophagus: implications for left atrial catheter ablation. Circulation 2004;110:3655–3660. 9. Tse HF, Lee KL, Fan K, et al. Nonfluoroscopic magnetic electroanatomic mapping to facilitate focal pulmonary veins ablation for paroxysmal atrial fibrillation. Pacing Clin Electrophysiol 2002;25:57– 61. 10. Tondo C, Mantica M, Russo G, et al. A new nonfluoroscopic navigation system to guide pulmonary vein isolation. Pacing Clin Electrophysiol 2005;28:S102– S105. 11. Rotter M, Takahashi Y, Sanders P, et al. Reduction of fluoroscopy exposure and procedure duration during ablation of atrial fibrillation using a novel anatomical navigation system. Eur Heart J 2005;26:1415–1421. 12. Dong J, Calkins H, Solomon SB, et al. Integrated electroanatomic mapping with three-dimensional computed tomographic images for real-time guided ablations. Circulation 2006;113:186 –194. 13. Kistler PM, Rajappan K, Jahngir M, et al. The impact of CT image integration into an electroanatomic mapping system on clinical outcomes of catheter ablation of atrial fibrillation. J Cardiovasc Electrophysiol 2006;17:1093–1101. 14. Martinek M, Nesser HJ, Aichinger J, et al. Impact of integration of multislice computed tomography imaging into three-dimensional electroanatomic mapping on clinical outcomes, safety, and efficacy using radiofrequency ablation for atrial fibrillation. Pacing Clin Electrophysiol 2007;30:1215–1223. 15. Dong J, Dickfeld T, Dalal D, et al. Initial experience in the use of integrated electroanatomic mapping with three-dimensional MR/CT images to guide catheter ablation of atrial fibrillation. J Cardiovasc Electrophysiol 2006;17:459 – 466. 16. Kistler PM, Earley MJ, Harris S, et al. Validation of three-dimensional cardiac image integration: use of integrated CT image into electroanatomic mapping system to perform catheter ablation of atrial fibrillation. J Cardiovasc Electrophysiol 2006;17:341–348. 17. Kuklik P, Szumowski L, Zebrowski JJ, et al. The reconstruction, from a set of points, and analysis of the interior surface of the heart chamber. Physiol Meas 2004;25:617– 627. 18. Kistler PM, Schilling RJ, Rajappan K, et al. Image integration for AF ablation: pearls and pitfalls. Heart Rhythm 2007;4:1216 –1221. 19. Zhong H, Lacomis JM, Schwartzman D. On the accuracy of CartoMerge for guiding posterior left atrial ablation in man. Heart Rhythm 2007;4:595– 602. 20. Fahmy TS, Mlcochova H, Wazni OM, et al. Intracardiac echo-guided image integration: optimizing strategies for registration. J Cardiovasc Electrophysiol 2007;18:276 –282.