Personalized Fluoroscopic Angles in Watchman™ Left Atrial Appendage Closure Landing Zone Assessment: A Three-Dimensional Printed Simulation Study

Background Atrial fibrillation causes ischemic stroke when thrombi dislodge from a cardiac outpouching, the left atrial appendage (LAA), and embolize to the brain. LAA occlusion with the Watchman™ device (Boston Scientific Corporation, MA, USA), which prevents stroke, requires accurate LAA measurements for device sizing. We explore whether standard fluoroscopic LAA measurements improve when obtained at CT-derived viewing angles personalized to LAA anatomy while concurrently referring to three-dimensional (3D) CT. Methods Left atrial 3D reconstructions created from contrast CT (n=28) were analysed to identify personalized viewing angles wherein LAA dimensions (LAA maximum landing zone diameter and LAA length) were best observed. The 3D-CT reconstructions were then 3D printed with stands. Fluoroscopy of anatomically oriented models in the catheter lab simulated LAA angiography. Fluoroscopic images were acquired at standard (caudal 20˚/right anterior oblique 30˚) and personalized viewing angles. Repeated measurements of LAA dimensions were taken from CT (Control), fluoroscopy at standard angles (Standard), personalized angles (Blinded), and personalized angles while concurrently referring to 3D CT (Referred). Results Control measurements correlated and agreed better with Referred and Blinded measurements than with Standard measurements (diameter correlation and agreement: Control/Standard r=.554, limits of agreement [LOAs]=6.83/-5.91; Control/Blinded r=.641, LOA =5.67/-5.54; Control/Referred r=.741, LOA=4.69/-4.14; length correlation and agreement: Control/Standard rs=.829, LOA=9.61/-3.02; Control/Blinded rs=0.789, LOA=7.13/-4.94; Control/Referred rs=.907, LOA=4.84/-4.13). Personalized angles resulted in hypothetical device size predictions more consistent with Control (device size correlation: Control/Standard rs=.698, Control/Blinded rs=.731, Control/Referred rs=.893, P<0.001). False ineligibility rates were Standard=6/28, Blinded=6/28, and Referred=2/28. Conclusion This simulation suggests that personalized fluoroscopic viewing angles with in-procedural reference to 3D CT may improve the accuracy of LAA maximum landing zone diameter and length measurements at the Watchman landing zone. This improvement may result in more consistent device size selection and procedural eligibility assessment. Further clinical research on these interventions is merited.


Conclusion
This simulation suggests that personalized fluoroscopic viewing angles with in-procedural reference to 3D CT may improve the accuracy of LAA maximum landing zone diameter and 1 1 1 1 1 1

Introduction
Atrial fibrillation (AF) substantially increases the risk of cardioembolic stroke [1]. In nonvalvular AF patients, stroke is often caused by thromboembolism from the left atrial appendage (LAA). Occluding the LAA in such patients can prevent stroke [2].
The Watchman™ device (Boston Scientific Corporation, MA, USA) is a commonly used transcatheter LAA occlusion device. Eligibility and device size selection for Watchman implantation depend on LAA dimensions, specifically, LAA landing zone diameter (measured from 1-2 cm anterior to the left upper pulmonary vein limbus to the inferior deflection of the LAA) and LAA maximum length along the implantation axis. LAA dimensions measured from fluoroscopy, contrast-enhanced cardiac CT, intracardiac echocardiography (ICE) and transesophageal echocardiography (TEE) are often discordant resulting in inaccurate device sizing [3]. The PROTECT AF trial required 1.8 device implantations per patient thus emphasizing the sizing dilemma [4].
CT LAA dimensions are obtained preprocedurally when the heart rate, rhythm and hydration status vary from the inprocedural state, resulting in non-representative results. Continuous TEE provides accurate and inprocedural dimensions, but is invasive and requires general anesthesia (GA) with associated risks. ICE is accurate, inprocedural and avoids GA; however, it is expensive, invasive and has limited views. Although fluoroscopic LAA dimensions are obtained inprocedurally and non-invasively, measurements are inaccurate compared to CT, TEE and ICE [3].
Fluoroscopic measurements of the anatomically variable LAA are taken at a standardized viewing angle, namely caudal 20˚ and right anterior oblique 30˚ (CAUD20/RAO30). CT-derived personalized fluoroscopic views have been proposed to improve measurement accuracy, yet have not been evaluated thus far [5].
Even with personalized views, the landing site selected on preprocedural CT may not be reselected on inprocedural fluoroscopy, perhaps because complex anatomy is poorly appreciated in two dimensions. We propose that inprocedural reference to three-dimensional (3D) CT during fluoroscopic analysis at personalized viewing angles can ensure measurements are made at comparable positions.
We investigate whether personalized fluoroscopic views and inprocedural reference to 3D CT can rectify fluoroscopic measurement errors. A simulation of LAA angiography using 3D printed left atrial (LA) models was developed due to the limited sample of Watchman cases in China.

Study protocol
A repeated measures experimental design was utilized. CT measurements of LAA dimensions, which accurately mirror model measurements, comprised the Control. Subsequently, three sets of fluoroscopic measurements were obtained, namely, LAA dimensions at (1) standard angles (Standard), (2) personalized angles (Blinded) and (3) personalized angles with concomitant reference to CT (Referred). LAA dimensions from each measurement category were used to assess eligibility for Watchman implantation, and to select a device size for implantation.
The correlation and agreement of the fluoroscopic measurements with Control measurements were explored. The clinical outcomes of the measurement techniques were investigated based on eligibility for Watchman implantation and predicted device size. Furthermore, contributors to measurement error (i.e., discrepancy between fluoroscopic measurements and Control measurements) were investigated using linear regression.  [6]. A customized stand oriented the LA identically to its CT position ( Figure 1). Models were saved in a stereolithographic (STL) format and post-processed on Blender 3D (Blender Foundation, Netherlands) to thicken model walls inwards (preserving external dimensions), ensure manifold meshes and remove free vertices. STL files were converted to G-Code (Slic3r 1.2.9, lead developer: Alessandro Ranellucci, Italy) and printed on a fused deposition modeling 3D printer (Prusa i3; eMotion Tech, France) in high-impact polystyrene plastic (Public Color, China) at a 0.3-mm layer resolution. The printer was calibrated using digital calipers to be accurate to 100 µm ensuring print-outs were an exact representation of the digital 3D-CT surface model. 3D, three dimensional.

Simulation of fluoroscopy
Models were oriented anatomically on the surgical table in a standard catheter laboratory (AXIOM Sensis XP; Siemens, Germany) with a pigtail catheter within the LAA (6F, Infiniti Angiographic Catheter; Cordis, CA, USA). Subsequently, they were examined under the fluoroscope in a coronary mode at Standard (CAUD20/RAO30) and personalized angles (described below). Fluoroscopic measurements of LAA maximum diameter and length were made on the AXIOM workstation (Siemens) using the pigtail catheter for calibration (Figures 2,  3).

Image analysis
Step 1: Identification of Watchman Landing Zone The area of the LAA at which the Watchman device would be expected to land was identified ( Figure 4) [7]. Step 2: Derivation of Personalized Fluoroscopic Viewing Angles From CT Figure 2 describes derivation of personalized fluoroscopic angles. At these angles, measurement points matching those on 3D CT can be measured on the fluoroscopic silhouette. Where feasible, a single fluoroscopic view for LAA maximum length and diameter was derived (n=8); otherwise two views optimized for either length or diameter measurements were derived (n=20) ( Figure 3).

Step 3: CT and Fluoroscopic Measurements at Landing Zone
The maximum LAA diameter and length at the landing zone were measured with CT and fluoroscopy as illustrated in Figure 4A and detailed below: 1. Control measurements obtained from CT represented true model dimensions. During analysis, the segmentation label map and 3D model were overlaid on multiplanar reconstruction CT slices. This two-dimensional (2D)-3D hybrid method ensured measurements made from CT were accurate representations of the 3D surface model ( Figure 2).
2. Standard measurements were obtained from model fluoroscopic images at a standard angle (CAUD20/RAO30).
3. Blinded measurements were obtained from model fluoroscopic images at personalized angles. During measurement, the operator was blinded from 3D CT. 4. Referred measurements were obtained on the reassessment of images at personalized fluoroscopic angles. Orthogonal 3D-CT views at personalized angles with the target landing zone marked were available for reference (Figures 2, 3).
Each fluoroscopic measurement was repeated three times with independent iterations of pigtail catheter calibration. Redundant measurements were averaged.

Eligibility
Landing zone measurements were analysed to assess eligibility for Watchman implantation. Eligibility criteria included LAA maximum length ≥LAA landing zone diameter, and LAA landing zone diameter ≥17 mm and ≤31 mm ( Figure 4) [8].

Device Size
If eligibility was established, a suitable device size (21, 24, 27, 30 or 33 mm) was selected with a target compression of 8%-20% at the LAA orifice ( Figure 4). If two device sizes were feasible, the larger size was selected. Details are presented in manufacturer-supplied directions for use [8].

Contributors to Fluoroscopic Measurement Errors
The following contributors to the fluoroscopic measurement error were assessed: LAA morphology (windsock, cactus, cone, bilobed, cauliflower or chicken wing) was recorded, and selected morphologies that may cause inaccurate measurement (cone, bilobed, cauliflower and chicken Wing) were grouped for analysis ( Figure 5) [9]. Eccentricity (1 -[short-axis LAA orifice diameter/long-axis diameter]) and absolute eccentricity (long-axis LAA orifice diameter -short-axis diameter) were also recorded ( Figure 6). The figure shows two oval LAA orifices of different sizes but identical shape casting fluoroscopic shadows (D, d) across their minimum landing zone diameter (dashed lines). When minimum diameter is measured instead of maximum diameter (solid lines) due to an inappropriate viewing angle, measurement error (minimum-maximum landing zone diameter) is greater in A than B.
Eccentricities (1 -[short-axis LAA orifice diameter/long-axis diameter]) in A and B are identical, and fail to predict a greater measurement error in A. Absolute eccentricity (long-axis diameter -shortaxis diameter) is greater in A than B, predicting greater measurement error in A.

Bias reduction
De-identified image datasets were analysed in a randomized order and independent observers were assigned to each measurement category. Operators were blinded to whether a fluoroscopic image was optimized for diameter or length, so both were measured for each image.

Statistical analysis
Sample size (n=28) was selected to detect a medium effect size with a power of 0.8 at P<.05 for all statistics described unless stated otherwise (G-power; Heinrich Heine University Düsseldorf, Germany). Continuous variables were described as means ± standard deviations, and categorical variables as frequencies.
Pearson's or Spearman's coefficient assessed correlation between continuous variables. Bland-Altman (BA) plots assessed agreement variables. Fixed bias was assessed with one sample t-test and proportional bias with linear regression.
Repeated measures analysis of variance (rANOVA), Friedman's ANOVA and Wilcoxon signedrank tests were used as appropriate. Multiple linear regression (forward step-wise) analyses was performed to assess predictors of CT maximum diameter and length. Assumptions of regression were assessed as appropriate.
P<0.05 was considered significant. For multiple post hoc comparisons, Bonferroni correction was applied, and reference P values were stated. All statistical tests were performed on IBM SPSS Statistics, Version 24 (IBM Corp, Armonk, NY).

Results
The average print time for a LAA model with the stand was 197 ± 48 minutes. The material cost (high-impact polystyrene) was 8.    Results of linear regression to predict the Control maximum diameter and maximum length are presented in Tables 2, 3. For Control maximum diameter, all assumptions of regression were met; however, for maximum length, mild heteroscedasticity was apparent.

Predictor variables
No. of cases (n) 28  2 Selected morphologies include cactus, cauliflower, chicken wing, cone and bilobed. 3 Step 2 for Referred measurements was excluded since selected morphologies did not add any significant predictive value.

Discussion
To the best of our knowledge, this simulation is the first to suggest improved accuracy of Watchman landing zone measurements through personalized fluoroscopic angles and inprocedural reference to 3D CT. Herein, we verify feasibility and hypothesize clinical benefit of these methods, justifying further clinical investigation.

Fluoroscopy in Watchman implantation
Fluoroscopy allows real-time, non-invasive and in-procedural imaging. However, in Watchman implantation, LAA fluoroscopic measurements are inaccurate compared to CT, TEE or ICE limiting its dependability for device sizing. While cardiac CT provides useful measurements, being preprocedural, it does not capture temporal changes in hydration status, heart rate and rhythm that can affect in-procedural LAA dimensions. While continuous TEE is an accurate alternative, it is invasive and requires GA. ICE is at par with TEE regarding measurement accuracy and clinical outcomes, is in-procedural and avoids GA; however, it is invasive, expensive and has limited views. Therefore, improving fluoroscopic measurements is of great importance.
Previous research demonstrated that CT-based personalized fluoroscopic angles benefit aortic valve measurements for TAVI [10]. Figure 8 illustrates how personalized angles can be beneficial. Herein, the same hypothesis is tested for LAA measurements. It is difficult to gauge intricate anatomy from the 2D fluoroscopic silhouette. 3D-CT registration on fluoroscopy during Watchman LAAC has been demonstrated to tackle this constraint [11]. Similarly, we referred to 3D-CT images during fluoroscopy, but while utilizing personalized viewing angles concurrently.
The clinical impact of measurement improvements is investigated through hypothetical device size selection and procedural eligibility assessment. Mechanisms by which interventions improve fluoroscopic measurements are investigated through regression analysis.

Simulations of Watchman implantation
Otton et al. first described Watchman device size prediction by implantation into a flexible 3D printed LAA model derived from CT [12]. Subsequent clinical studies have established this method in device sizing [13]. A 3D printed phantom simulation of LAA fluoroscopic angiography, established previously by Sra et al., is utilized herein to test interventions to improve fluoroscopic measurement accuracy [14].
This simulation demonstrates personalized fluoroscopic viewing angles and reference to 3D-CT images during fluoroscopy (Referred measurement technique) improves fluoroscopic measurement accuracy (i.e. increased correlation and narrowed limits of agreement of fluoroscopic LAA measurements with Control). Incremental improvements support concomitant use of both interventions (Figure 7, Tables 2, 3).
Underestimating LAA landing zone diameter results in device undersizing leading to reduced stability and possible device embolization. The Referred measurement technique was less likely to underestimate the LAA landing zone diameter and may speculatively reduce device embolization (represented by a 2.14-mm narrowing in the upper LOA between Standard and Referred measurements of maximum diameter) (Figure 7). Furthermore, the increased measurement accuracy with Referred measurements resulted in more consistent device size selection with Control (device size correlation: Control/Referred rs=.893 vs. Control/Standard rs=.695, P<0.001).
If the LAA length cannot accommodate the length of the Watchman device, a patient is considered ineligible for implantation. Underestimation of maximum length was reduced with Referred measurements as compared to the Standard technique, represented by a 4.77-mm narrowing in the upper LOA between the two. This may prevent patients from incorrectly being excluded from Watchman LAAC due to inaccurate length measurements (false exclusions: Standard=6/28, Blinded=6/28, Referred=2/28) [8].
In our sample, improvements in device size prediction and reductions in false exclusions support replacement of the Standard measurement technique with the Referred technique. Improvements in length measurements, but not diameter measurements are generalizable based on 95% confidence intervals of LOA (Figure 7) [15].

Contributors to measurement error
Through linear regression, contributors to fluoroscopic measurement error and how interventions mitigated error were investigated.
The contribution of LAA landing zone eccentricity and absolute eccentricity to maximum diameter measurement error was explored. In our sample, eccentricity did not contribute to measurement error. Speculatively, while appendages differing greatly in size may share the same eccentricity, the maximal measurement error due to fluoroscopic views would be greater for a larger appendage ( Figure 6). with additional reference to 3D CT (Referred ΔR 2 =11%) was unrelated to absolute eccentricity and was possibly attributable to the consistent selection of measurement points between 3D CT and fluoroscopy ( Table 2).
The 2D visualization of LAA morphology complicates accurate fluoroscopic measurements by making selection of implantation location, direction and strategy difficult. Selected morphologies that are challenging to measurements are described in Figure 5. Selected morphologies contributed equally to measurement errors for Standard and Blinded measurements. The Referred measurement technique mitigated morphology-induced error, perhaps by improving anatomical appreciation of the 2D LAA silhouette ( Table 3).

Limitations
This study is limited as it is a simulation, and consecutive sampling was used. Simulation allowed us to evaluate feasibility, provide proof of concept and hypothesize benefit despite a limited sample of Watchman implantations in China.
Differences between this simulation and a clinical setting merit consideration. CT measurements herein closely mirror 3D printed model measurements (through printer calibration), and are the Control for comparisons with fluoroscopic measurements.
Herein, cardiac CT was acquired at 78% R-R (for compatibility with the Carto XP system; Biosense Webster, Inc., CA, USA), while ventricular systole captures largest LAA dimensions. The simulated fluoroscopic silhouette is assumed to represent maximum LAA dimensions, and systematic error due to cardiac phase is rectified by the repeated measures design. Clinically, imaging a standardized cardiac phase between modalities is more important since a dynamic heart is imaged rather than a static model.
As this was a simulation, clinical variables are not available. Comparisons between CT, TEE and personalized fluoroscopic measurements and their effects on device sizes selected, percentage of successful implantations and number of devices deployed require further investigation.

Conclusions
The results of this simulation study of LAA angiography in Watchman device implantation show that personalized fluoroscopic viewing angles and in-procedural reference to 3D CT may incrementally improve the accuracy of fluoroscopic measurements of LAA dimensions, as compared to standard fluoroscopic angles. Improved measurement accuracy will result in improved consistency in device size predictions and procedural eligibility assessment. Based on this evidence, further clinical evaluation of these methods is indicated.

Additional Information Disclosures
Human subjects: Consent was obtained by all participants in this study. Animal subjects: All authors have confirmed that this study did not involve animal subjects or tissue. Project (LC2016ZD002). Financial relationships: All authors have declared that they have no financial relationships at present or within the previous three years with any organizations that might have an interest in the submitted work. Other relationships: All authors have declared that there are no other relationships or activities that could appear to have influenced the submitted work.