Feasibility and accuracy of automated three-dimensional echocardiographic analysis of left atrial appendage for transcatheter closure

Page 1/33 Title: Feasibility and Accuracy of Automated Three-Dimensional Echocardiographic Analysis of Left Atrial Appendage for Transcatheter Closure Pedro Morais, PhDa,*, Yiting Fan, MMb,*, Sandro Queirós, PhDc,d, Jan D’hooge, PhDe, Alex PuiWai Lee, MDf,g,#, João L. Vilaça, PhDa,# a2Ai – School of Technology, IPCA, Barcelos, Portugal b Cardiology Department, Shanghai Chest Hospital, Shanghai Jiaotong University, China cLife and Health Sciences Research Institute (ICVS), School of Medicine, University of Minho, Braga, Portugal dICVS/3B’s-PT Government Associate Laboratory, Braga/Guimarães, Portugal. eLab on Cardiovascular Imaging & Dynamics, Department of Cardiovascular Sciences, KULeuven - University of Leuven, Leuven, Belgium fDivision of Cardiology, Department of Medicine & Therapeutics, Prince of Wales Hospital, Hong Kong, China; gLaboratory for Cardiac Imaging and 3D Printing, Li Ka Shing Institute of Health Science, Faculty of Medicine, The Chinese University of Hong Kong, Hong Kong, China * - joint first authors contributed equally # - joint last authors contributed equally Page 2/33 Corresponding author: Alex Pui-Wai Lee, E-mail: [email protected]k Laboratory for Cardiac Imaging and 3D Printing, Li Ka Shing Institute of Health Science, Faculty of Medicine, The Chinese University of Hong Kong, Hong Kong, China Conflicts of Interest: Nothing to declare Funding This work was funded by the projects “NORTE-01-0145-FEDER-000045” and “NORTE-010145-FEDER-000059”, supported by Northern Portugal Regional Operational Programme (Norte2020), under the Portugal 2020 Partnership Agreement, through the European Regional Development Fund (FEDER). It was also funded by national funds, through the FCT – Fundação para a Ciência e Tecnologia and FCT/MCTES in the scope of the project UIDB/05549/2020 and UIDP/05549/2020 and the grant CEECIND/03064/2018. This work was also funded by the Hong Kong Special Administrative Region Government Health and Medical Research Fund (05160976). Highlights - Accurate device sizing is of paramount importance for procedural success of LAAC - A software for automated device sizing in 3D-TEE is assessed in a retrospective cohort. - Automated device sizing achieved the highest success rate with 90.5% accuracy. - Automated device sizing is reproducible and saves time for LAAC analysis. Page 3/33 Abstract Background: Procedural success of transcatheter left atrial appendage closure (LAAC) is dependent on correct device selection. Three-dimensional transesophageal echocardiography (3DTEE) is more accurate than the two-dimensional (2D) modality for evaluation of the complex anatomy of LAA. However, 3D-TEE analysis of LAA is challenging and highly expertise dependent. In this study, we sought to evaluate the feasibility and accuracy of a novel software tool for automated 3D analysis of the LAA using 3D-TEE data. Methods: The intra-procedural 3D TEE data of 158 patients who underwent LAAC were retrospectively analyzed with a novel automated LAA analysis software tool. Based on the 3D TEE data, the software semi-automatically segmented the 3D LAA structure, determined the device landing zone (LZ), and generated measurements of the LZ dimensions and the LAA length, allowing manual editing if necessary. The accuracy of LAA pre-implantation anatomic measurement reproducibility, and time for analysis of the automated software were compared against expert manual 3D analysis. The software feasibility to predict the optimal device size was directly compared to implanted models. Results: Automated 3D analysis of the LAA on 3D-TEE was feasible in all patients. There were excellent agreements between automated and manual measurements of LZ maximal diameter (bias:-0.32, LOA:-3.56; 2.92), area-derived mean diameter (bias:-0.24, LOA:-3.12, 2.64), and LAA depth (bias:0.02, LOA:-3.14; 3.18). Automated 3D analysis, with manual editing if necessary, accurately identified the implanted device size in 90.5% of patients, outperforming 2DTEE (68.9%, p<0.01). The automated software showed results competitive against the manual analysis of 3D-TEE with higher intraand inter-observer reproducibility and allowed quicker analysis (101.9±9.3s vs. 183.5±42.7s, p<0.001) compared to manual analysis. Page 4/33 Conclusions: Automated LAA analysis based on 3D-TEE is feasible and allows accurate, reproducible and rapid device sizing decision for LAAC. Keywords: Left Atrial Appendage Closure, LAA device sizing, Automated software analysis; Interactive analysis; 3D transesophageal echocardiography. Abbreviations: ACP – Amulet Cardiac Plug CA – Circumflex Artery CT – Computed Tomography FOV – Field of view ICC - Intraclass correlation coefficient LAA – Left Atrial Appendage LAAC – Left Atrial Appendage Closure LZ – Landing Zone MPR – Multiplanar reconstruction TEE – Transesophageal Echocardiography WM – Watchman Page 5/33 Introduction Occluder device sizing for transcatheter left atrial appendage closure (LAAC) is a challenging task requiring careful evaluation of the highly variable LAA anatomy on periprocedural transesophageal echocardiography (TEE), fluoroscopy, or computed tomography (CT)1. Suboptimal device sizing has direct impact on the procedural time, safety and efficacy 1,2. Device underor over-sizing may result in complications including device embolization, device-related thrombus, peri-device leak, prolonged procedural time, or cardiac tamponade3. Although twodimensional (2D) TEE is currently the standard imaging modality for LAAC device sizing, it frequently underestimates LAA orifice size owing to foreshortening4, especially in cases where the orifices are oval-shaped. Three-dimensional (3D) TEE has clear advantages over 2D TEE in the evaluation of LAA anatomy1,3–5. Volume rendering of the 3D TEE image allows en face visualization of the ellipticity of LAA orifice; moreover, through multiplanar reconstruction (MPR)3,4,6–8 the 3D volume can be reformatted to multiple 2D views (standard or modified) for measurements of LAA diameters and depths without foreshortening; additionally, MPR of the short-axis plane of LAA allows measurement of the area and other 3D-derived parameters of the device landing zone (LZ), which have been shown to be more useful for occluder device sizing compared to conventional 2D parameters5. Nevertheless, 3D TEE has not yet been widely adopted in clinical practice for periprocedural LAAC device sizing, with several possible explanations: (i) 3D navigation using orthogonal planes is not straightforward, has to be done manually, requiring specific training and software; (ii) 3D TEE image analysis can be time-consuming, which may hamper its implementation in intraprocedural analysis; and (iii) interpretation and measurement of the LAA 3D anatomy, including the identification of device LZ plane, is highly expertise-dependent with relatively Page 6/33 reduced observer reproducibility. Automated analysis of 3D echocardiographic data has shown promise in overcoming these limitations in other scenarios such as chamber quantification and valve analysis 9–11. For LAA, automated segmentation in CT using image-based methods12 and deep learning13 were proposed. However, TEE remains the most frequently used modality for preprocedural LAA imaging and is the only viable option that can provide 3D intra-procedural image guidance. Our group has developed an automated modeling algorithm that allows rapid segmentation of LAA volume data obtained by 3D TEE14 with automatic extraction of 3D LAA geometry from the segmented models15. Our methods were validated in a small number of datasets (~20 cases), demonstrating its feasibility in terms of 3D segmentation and 3D-based geometric measurements. In the present study, we sought to evaluate the diagnostic performance and reproducibility of the proposed automated 3D-TEE-based method for LAAC device sizing in comparison to the conventional manual approach in a larger patient cohort undergoing LAAC using two commonly used devices. Methods Study Population A total of 183 AF patients who underwent LAAC at the Prince of Wales Hospital (Hong Kong) using WatchmanTM (WM, n=112; Boston Scientific, MA, USA) and AmplatzerTM Cardiac Plug 2/AmuletTM (ACP, n=71; Abbott, IL, USA), were retrospectively reviewed. All subjects had nonvalvular AF, CHA2DS2–VASc score ≥1, and relative or absolute contraindications to longterm anticoagulation. Exclusion criteria were patients with no intra-procedural full cardiac-cycle 2D/3D TEE sequence available (n=20) and the impossibly to export TEE data for an open format (n=5). Finally, a total of 158 cases (98 WM and 60 ACP) were selected for the study. The study received the approval of the Ethics Committee of the Prince of Wales Hospital (Code 2020.139). Page 7/33 LAAC and Procedural Imaging LAAC was performed using a transfemoral transseptal approach under monitored anesthesia care and guidance by 3D TEE and fluoroscopy in all patients. The transseptal puncture was guided using matrix-array biplane mode in the bicaval (90°) and short-axis (45°) views. Intraprocedural 2D TEE images of the LAA were acquired with transducer array rotated through 0°, 45°, 90°, and 135° in the mid-esophageal position16 after raising LA pressure to ≥10 mm Hg with intravenous normal saline. A team of expert interventionists and imagers assessed the acquired images and performed manual measurements of the device landing zone (LZ) and the LAA depth. For WM, LZ diameters from the circumflex artery (CA) inferiorly to a point superiorly 1–2 cm within the pulmonary vein ridge were recorded by sweeping through the standard imaging planes from 0° to 135°. The depth of LAA was subsequently measured from the orifice LZ to the tip of the deepest LAA lobe 16. For ACP, LZ diameters were measured 1−1.5 cm distal to a line connecting the CA to the tip of the pulmonary vein ridge. The LAA depth was calculated as the distance of a perpendicular line from the LAA orifice to the roof of the appendage16. An illustration scheme of each sizing strategy is presented in Supplementary Material A. 3D TEE is the standard procedural imaging technique for LAAC at our center17. All patients underwent intraprocedural 3D TEE with a Philips EPIQ 7, CVx or IE33 ultrasound machine (Philips, Bothell, WA). Image acquisition was performed in the mid-esophageal position using the “3D zoom” mode, keeping the LAA body, CA, left pulmonary vein ridge, and part of LA and mitral valve in the field of view (FOV)2. The average frame rate of the 3D images used in this study was 15 Hz ± 9 Hz (~12 ± 7 frames per cardiac cycle). The resulting 3D image presented a resolution and size that varied from 0.22 to 0.86 mm and 112 × 112 × 208 to 336 × 320 × 224 voxels, respectively. The 3D datasets of LAA were assessed manually on-cart using the Page 8/33 multiplanar reconstruction (MPR) mode, which allowed manual alignment of the LAA in 3 orthogonal planes (2 long-axis, 1 short-axis). The LAA long-axis planes were aligned, and the short-axis plane was adjusted to visualize the device LZ plane. The area, maximum, and minimum diameter of the LZ were manually measured on the short-axis plane, and the depth was measured from the long-axis views. Moreover, the LAA morphology was classified as described in Di Biase et al.18. Based on the obtained measurements, patients underwent implantation of the WM and ACP devices. WM is available in 5 different devices - 21, 24, 27, 30 and 33 sizes - while ACP has 8 different devices - 16, 18, 20, 22, 25, 28, 31 and 34 sizes. For each patient, device sizing at the time of implantation was made at the operator’s discretion, using data from all available modalities (2D and 3D TEE and fluoroscopy), taking into account the LZ diameters and LAA depth1,3,7,16,19, as well as other factors including LAA morphology and trabeculae, particularly in borderline situations (e.g. measurements between two consecutive sizes in the manufacturer’s guidelines)17. After implantation, the device was evaluated to confirm correct sizing, position, and stability, as previously described1,2. Peri-device leaks were assessed using Doppler echocardiography. Automated LAA Software Blinded to the procedural data, automated analysis of the 3D TEE datasets was performed retrospectively on the intraprocedural 3D TEE datasets using the LAA software tool integrated on a customized non-commercial MATLAB-based application (MathWorks, Natick, MA), termed Speqle3D (KULeuven, Leuven, Belgium)11. The data were exported in Cartesian DICOM format and converted to an isotropic voxel spacing for evaluation using the proposed software. For each patient, only the sequence before the implantation moment was evaluated by the proposed software. In this study, two automated operating modes based on the proposed LAA software were Page 9/33 assessed, namely: automated results without manual editing (henceforward referred Auto) and automated analysis followed by manual editing, if necessary (henceforward mentioned Interactive). The software workflow is divided into three stages (Figure 1): (1) identification of the LAA centerline, through three or more points defined manually in the LAA anatomy, and the CA, based on one point or a 2D contour (Figure 1-A); (2) 3D segmentation of the LAA lumen boundaries (Figure 1-B)14; and (3) estimation of all required measurements for device sizing, i.e. device LZ short-axis level and LAA depth distance (Figure 1-C). While manual initialization is required in stage 1, the remaining stages of the software tool are automatic15. The operator starts by selecting the end-systolic frame with maximal LAA volume in the cardiac cycle. By navigation through 3 orthogonal reconstructed planes of the 3D volume, three points are manually located at the proximal, central, and distal parts (tip) of LAA, which are then interpolated for a 3D centerline, allowing manual corrections if required. Then, a specific navigation mode through a set of orthogonal planes (i.e. one short-axis and two long-axis) defined along the centerline (i.e. shortaxis plane’s normal is aligned with the LAA centerline) is automatically activated. The centerlinebased navigation ensures that 2D views are centered and co-axial with the proximal LAA, simplifying the study of complex morphology particularly in high curvature portions. The approximate location of CA is then manually defined (Stage 1). The above manual steps initialize a mean 3D LAA deformable model14 that evolves to fit the patient-specific LAA anatomy using intensity-based and edge-based terms, segmenting the LAA lumen (Stage 2). Finally, the LAA device LZ short-axis plane is computed automatically by the software through alignment of the 3D segmented surface with a set of known 3D references through a weighted iterative closest point approach15. This alignment step is repeated with independent device-specific templates to estimate Page 16/33 However, 2D TEE often underestimates LAA dimensions compared to 3D modalities, including CT and 3D TEE. 3D imaging allows reconstruction of anatomically correct planes allowing measurements of LAA dimensions without foreshortening. Nevertheless, manual reconstruction of the LAA 3D datasets is time-consuming; this is particularly undesirable for intraprocedural analysis when decisions must be made rather quickly to avoid prolonging the procedural time. Furthermore, the LAA anatomy is highly complex, and deciding the device implantation level can be difficult, with high observer variability. The present study involved a larger cohort of patients undergoing LAAC using two commonly used devices and provides new insights into the clinical utility of automated LAA analysis by demonstrating a high accuracy in correct device size identification and efficiency, as well as reproducibility in LAA anatomical measurements. Additionally, the automated software achieved similar performance for all LAA morphologies. Nevertheless, we must mention that a slightly inferior performance (not statistically significant) was found in morphologies with high anatomical heterogeneity at the distal portion (e.g., large pectinate muscles), most common in cactus and cauliflower morphologies. In contrast, the manual analysis was more challenging in high curvature cases (i.e., chicken wing and windsock). Nevertheless, these results are still preliminary, requiring further studies with a similar representation of all LAA shapes. To measure the feasibility of the proposed software in terms of correct identification of the LAA device, multiple experimental scenarios were defined and compared offline with retrospective data from real interventions. Since the LAA is a fluid mediator with the capability to change its dimensions4, a variation of its size with the increase in LA pressure is expected which may result in differences between preprocedural and intraprocedural clinical indexes. Thus, all analyses of this study were performed based on intra-operative TEE images. Nevertheless, the application of Page 17/33 the proposed software in other interventional stages (namely pre-procedural 3D TEE imaging) is also viable. Many investigators have explored the clinical value of preprocedural measurements by using CT-based measurements 25,26. Different strategies have been assessed to minimize the influence of possible LAA anatomical expansion throughout the intervention 25,26. This typically is performed by upsizing of the preprocedural measurements through a pre-defined rule3,26. Nevertheless, since the increase of the LAA diameter between preand intraprocedural stages varies between subjects 27,28, the identification of the optimal upsizing value is still highly dependent on the operator which makes it suboptimal for clinical practice. Thus, in our clinical practice, when CT data is pre-procedurally acquired, intraoperative 3D TEE images are still mandatory to confirm the anatomy and the device sizing at the implantation moment, making the proposed automated 3D software of high clinical interest. In this study, all image volumes were acquired by expert echocardiographers in their normal clinical routine and were of high quality; whether our automated approach applies to poorer quality images acquired by non-expert or how much its performance depends on operator expertise needs further evaluation. Our preliminary analysis (see supplementary material E) suggests that the operator’s expertise does not have a high impact on the performance of the automated software (i.e. Auto mode), with experts and average users achieving a similar success rates in terms of device prediction. The interactive and auto modes performed by an average user proved feasible, achieving a success rate higher than manual analysis by the nonexpert. In contrast, for 3D manual analysis, the expert clearly outperformed the average user. The current software is versatile and allows automated and reproducible computation of the area and perimeter of device LZ, which may be more accurate for device sizing than maximal LZ diameters, as suggested by previous studies4,28. Furthermore, it allows fast automatic 3D Page 18/33 segmentation of the LAA, which may potentiate the development of quicker 3D printing techniques2. 4D analysis (i.e. 3D segmentation plus tracking) is currently under investigation and may provide clinically relevant functional information of the LAA29. Study limitations Our study has limitations. First, it was a single-center and single vendor retrospective study. Nevertheless, the study cohort is relatively large, with a variety of LAA sizes and morphologies. Moreover, the 3D TEE images tested in this study were prospectively acquired following a standard LAAC imaging protocol17. Although the current study relies on 3D TEE datasets from a single vendor, the software was already tested in volumes from a different vendor (see our previous studies14,15) maintaining its high performance and feasibility. Second, only the WM and ACP occluders were used, while there are other devices available in the market. However, WM and ACP represent the two most frequently used endocardial LAAC devices adopting two different types of designs (i.e. “parachute” and “lobe and disk”), supporting the applicability of our software in common practice. It might be noted that the application of the proposed software to other devices (e.g. Lambre) is also possible, simply requiring the generation of a set of 3D reference anatomical models with the values of the manufacturer-recommended target clinical levels for device sizing. Third, our automated software was applied to only 3D TEE images; we did not test the feasibility of automated analysis on CT images, which is used in some centers for preprocedural planning3,23. Nonetheless, we believe applying automated LAA analysis on TEE is clinically more relevant than applying it on CT because the former is the standard imaging modality used intraprocedurally. Although CT provides excellent image quality for accurate preprocedural LAA evaluation, the final device sizing decision is often made intraprocedurally after volume loading28. Fourth, although we benchmarked against the clinically implanted device size, this is not necessarily the Page 19/33 best or only suitable size. Multiple factors may influence the implantation result, namely modification of the implantation site in specific anatomies or oversizing/undersizing of the extracted measurement based on the operator experience. In light of this, the reported accuracy (~90%) is extremely good and higher accuracy is unlikely. Moreover, due to all these sources of potential variability, the comparison between the automated software and manual analysis in terms of device size prediction must be considered a proof of concept of the potential clinical value of our software and not as a direct comparison between different sizing strategies. Finally, our approach remains semi-automatic requiring initial manual definition of the LAA centerline. A fully automatic approach could further reduce processing time; however, we believe our semi-automatic interactive approach is suitable for clinical use, allowing expert input in device sizing, with demonstrable superior performance and efficiency compared to the traditional fully manual practice. Conclusions This study provides validation of a novel 3D software for automated LAAC device sizing using 3D TEE images. The described software proved to be feasible, accurate, and reproducible. The interactive approach, with automation and manual editing when necessary, proved to be a potentially useful clinical option, showing a performance comparable to the standard manual practice. Overall, the obtained results indicate that automated 3D analysis of the LAA with the proposed software has the potential to be integrated into clinical practice, promoting routine use of 3D TEE for procedural guidance of LAAC. References Page 20/33 1. Glikson M, Wolff R, Hindricks G, Mandrola J, Camm AJ, Lip GYH, et al. EHRA/EAPCI expert consensus statement on catheter-based left atrial appendage occlusion--an update. EP Eur. 2020;22(2):184. 2. Fan Y, Yang F, Cheung GS-H, Chan AK-Y, Wang DD, Lam Y-Y, et al. Device sizing guided by echocardiography-based three-dimensional printing is associated with superior outcome after percutaneous left atrial appendage occlusion. J Am Soc Echocardiogr. 2019;32(6):708–19. 3. Saw J, Fahmy P, Spencer R, Prakash R, McLaughlin P, Nicolaou S, et al. Comparing measurements of CT angiography, TEE, and fluoroscopy of the left atrial appendage for percutaneous closure. J Cardiovasc Electrophysiol. 2016;27(4):414–22. 4. Freixa X, Aminian A, Tzikas A, Saw J, Nielsen-Kudsk J-E, Ghanem A, et al. Left atrial appendage occlusion with the amplatzer amulet: update on device sizing. J Interv Card Electrophysiol. 2020;1–8. 5. Zhou Q, Song H, Zhang L, Deng Q, Chen J, Hu B, et al. Roles of real-time threedimensional transesophageal echocardiography in peri-operation of transcatheter left atrial appendage closure. Medicine (Baltimore). 2017;96(4). 6. Jia D, Zhou Q, Song H, Zhang L, Chen J, Liu Y, et al. The value of the left atrial appendage orifice perimeter of 3D model based on 3D TEE data in the choice of device size of LAmbreTM occluder. Int J Cardiovasc Imaging. 2019;35(10):1841–51. 7. Schmidt-Salzmann M, Meincke F, Kreidel F, Spangenberg T, Ghanem A, Kuck K-H, et al. Improved algorithm for ostium size assessment in watchman left atrial appendage occlusion using three-dimensional echocardiography. J Invasive Cardiol. 2017;29(7):232–8. Page 21/33 8. Korsholm K, Berti S, Iriart X, Saw J, Wang DD, Cochet H, et al. Expert recommendations on cardiac computed tomography for planning transcatheter left atrial appendage occlusion. Cardiovasc Interv. 2020;13(3):277–92. 9. Jin C-N, Salgo IS, Schneider RJ, Kam KK-H, Chi W-K, So C-Y, et al. Using anatomic intelligence to localize mitral valve prolapse on three-dimensional echocardiography. J Am Soc Echocardiogr. 2016;29(10):938–45. 10. Luo X-X, Fang F, So H-K, Liu C, Yam M-C, Lee AP-W. Automated left heart chamber volumetric assessment using three-dimensional echocardiography in Chinese adolescents. Echo Res Pract. 2017;4(4):53–61. 11. Queirós S, Morais P, Dubois C, Voigt J-U, Fehske W, Kuhn A, et al. Validation of a novel software tool for automatic aortic annular sizing in three-dimensional transesophageal echocardiographic images. J Am Soc Echocardiogr. 2018;31(4):515–25. 12. Leventić H, Babin D, Velicki L, Devos D, Galić I, Zlokolica V, et al. Left atrial appendage segmentation from 3D CCTA images for occluder placement procedure. Comput Biol Med. 2019;104:163–74. 13. Jin C, Feng J, Wang L, Yu H, Liu J, Lu J, et al. Left atrial appendage segmentation using fully convolutional neural networks and modified three-dimensional conditional random fields. IEEE J Biomed Heal informatics. 2018;22(6):1906–16. 14. Morais P, Queirós S, De Meester P, Budts W, Vilaça JL, Tavares JMRS, et al. Fast segmentation of the left atrial appendage in 3-D transesophageal echocardiographic images. IEEE Trans Ultrason Ferroelectr Freq Control. 2018;65(12):2332–42. Page 22/33 15. Morais P, Vilaça JL, Queirós S, De Meester P, Budts W, Tavares JMRS, et al. Semiautomatic Estimation of Device Size for Left Atrial Appendage Occlusion in 3-D TEE Images. IEEE Trans Ultrason Ferroelectr Freq Control. 2019 May;66(5):922–9. 16. Yu C-M, Khattab AA, Bertog SC, Lee APW, Kwong JSW, Sievert H, et al. Mechanical antithrombotic intervention by LAA occlusion in atrial fibrillation. Nat Rev Cardiol. 2013;10(12):707. 17. So C, Cheung GS, Chan AK, Lee AP, Lam Y. A call for standardization in left atrial appendage occlusion. J Am Coll Cardiol. 2018;72(4):472–3. 18. Di Biase L, Santangeli P, Anselmino M, Mohanty P, Salvetti I, Gili S, et al. Does the left atrial appendage morphology correlate with the risk of stroke in patients with atrial fibrillation? Results from a multicenter study. J Am Coll Cardiol. 2012;60(6):531–8. 19. Saw J, Lempereur M. Percutaneous left atrial appendage closure: procedural techniques and outcomes. JACC Cardiovasc Interv. 2014;7(11):1205–20. 20. Chow DHF, Bieliauskas G, Sawaya FJ, Millan-Iturbe O, Kofoed KF, Søndergaard L, et al. A comparative study of different imaging modalities for successful percutaneous left atrial appendage closure. Open Hear. 2017;4(2). 21. Nucifora G, Faletra FF, Regoli F, Pasotti E, Pedrazzini G, Moccetti T, et al. Evaluation of the left atrial appendage with real-time 3-dimensional transesophageal echocardiography: implications for catheter-based left atrial appendage closure. Circ Cardiovasc Imaging. 2011;4(5):514–23. 22. Mediratta A, Addetia K, Medvedofsky D, Schneider RJ, Kruse E, Shah AP, et al. 3D Page 23/33 echocardiographic analysis of aortic annulus for transcatheter aortic valve replacement using novel aortic valve quantification software: Comparison with computed tomography. Echocardiography. 2017;34(5):690–9. 23. Xu B, Betancor J, Sato K, Harb S, Rehman KA, Patel K, et al. Computed tomography measurement of the left atrial appendage for optimal sizing of the Watchman device. J Cardiovasc Comput Tomogr. 2018;12(1):50–5. 24. Rajwani A, Nelson AJ, Shirazi MG, Disney PJS, Teo KSL, Wong DTL, et al. CT sizing for left atrial appendage closure is associated with favourable outcomes for procedural safety. Eur Hear Journal-Cardiovascular Imaging. 2017;18(12):1361–8. 25. Nadeem F, Igwe C, Stoycos S, Jaswaney R, Tsushima T, Al-Kindi S, et al. A new WATCHMAN sizing algorithm utilizing cardiac CTA. Cardiovasc Revascularization Med. 2021; 26. Roy AK, Horvilleur J, Cormier B, Cazalas M, Fernandez L, Patane M, et al. Novel integrated 3D multidetector computed tomography and fluoroscopy fusion for left atrial appendage occlusion procedures. Catheter Cardiovasc Interv. 2018;91(2):322–9. 27. Spencer RJ, DeJong P, Fahmy P, Lempereur M, Tsang MYC, Gin KG, et al. Changes in left atrial appendage dimensions following volume loading during percutaneous left atrial appendage closure. JACC Cardiovasc Interv. 2015;8(15):1935–41. 28. Al-Kassou B, Tzikas A, Stock F, Neikes F, Völz A, Omran H. A comparison of twodimensional and real-time 3D transoesophageal echocardiography and angiography for assessing the left atrial appendage anatomy for sizing a left atrial appendage occlusion system: impact of volume loading. EuroIntervention J Eur Collab with Work Gr Interv Page 24/33 Cardiol Eur Soc Cardiol. 2017;12(17):2083–91. 29. Schluchter A, Jan C, Lowe K, Vigneault DM, Contijoch F, McVeigh ER. Vascular Landmark-Based Method for Highly Reproducible Measurement of Left Atrial Appendage Volume in Computed Tomography. Circ Cardiovasc Imaging. 2019;12(12):e009075. Page 25/33 Figures: Figure 1 – Overview of the automated LAA software. After (A) initial manual definition of the LAA centerline, the automated software (B) three-dimensionally segments the LAA, and (C) automatically locates the device landing zone for automated anatomic measurements relevant to device sizing. The cyan plane is the LZ. The white dotted line represents the LAA depth. CA – Circumflex artery. Page 32/33 Table 1 - Patients’ baseline characteristics Parameter Value Age, years 73.5±7.8 Men, n(%) 104 (66%) LAA Morphology, n(%) Chicken Wing 24 (15%) Windsock 23 (15%) Cauliflower 81 (51%) Cactus 30 (19%) Medical History Paroxysmal AF, n(%) 63 (40%) Non-paroxysmal AF, n(%) 95 (60%) CHA2DS2-VASc score 4.4±1.5 1-3, n(%) 43 (27%) 4-6, n(%) 103 (65%) 7-9, n(%) 12 (8%) HAS-BLED score 3.0±1.0 1-2, n(%) 42 (27%) 3-4, n(%) 107 (68%) 5-6, n(%) 9 (6%) LV Ejection Fraction, % 57.7±9.0 Reduced (<40%), n(%) 8 (5%) Mid-range (41%-49%), n(%) 6 (4%) Normal (50%-70%), n(%) 142 (90%) High (>70%), n(%) 2 (1%) Page 33/33 Table 2 – Intraand inter-observer variability for LZ area-derived mean diameters, maximum diameters, and LAA depth of automated (Auto and Interactive modes) and manual analyses WM ACP Intra Inter Intra Inter ICC (95% CI) ICC (95% CI) ICC (95% CI) ICC (95% CI) Mean LZ diameter Auto 0.93 (0.89-0.95) 0.82 (0.78-0.89) 0.91 (0.85-0.95) 0.86 (0.82-0.91) Interactive 0.93 (0.90-0.96) 0.85 (0.80–0.91) 0.92 (0.86-0.95) 0.84 (0.80-0.90) 3D manual analysis 0.90 (0.87-0.94) 0.80 (0.70-0.88) 0.90 (0.72-0.95) 0.78 (0.68-0.87) Maximum LZ diameter Auto 0.90 (0.84-0.94) 0.82 (0.77–0.88) 0.84 (0.81-0.90) 0.78 (0.74-0.85) Interactive 0.92 (0.90-0.95) 0.80 (0.75-0.86) 0.88 (0.84-0.93) 0.80 (0.74-0.86) 3D manual analysis 0.86 (0.80-0.90) 0.76 (0.69-0.84) 0.80 (0.71-0.88) 0.71 (0.61-0.81) LAA Depth Auto 0.82 (0.74-0.88) 0.80 (0.71-0.85) 0.85 (0.72-0.90) 0.80 (0.68-0.90) Interactive 0.84 (0.75-0.91) 0.76 (0.69-0.83) 0.86 (0.73-0.92) 0.78(0.73-0.82) 3D manual analysis 0.75 (0.65-0.83) 0.70 (0.65–0.76) 0.84 (0.75-0.87) 0.67 (0.60-0.74)