Feasibility, Accuracy, and Reproducibility of Aortic Annular and Root Sizing for Transcatheter Aortic Valve Replacement Using Novel Automated Three-Dimensional Echocardiographic Software: Comparison with Multi–Detector Row Computed Tomography


In transcatheter aortic valve replacement (TAVR), multi–detector row computed tomography (MDCT) is currently the standard imaging modality for correct prosthesis sizing, despite risks of radiation and contrast-induced renal injury. Three-dimensional (3D) transesophageal echocardiography (TEE) has been proposed as a potential alternative imaging technique, and recently, automated 3D transesophageal echocardiographic software (Aortic Valve Navigator [AVN], an unreleased prototype from Philips) has been developed for assessment of the aortic annulus and root. The aim of this study was to assess the feasibility, accuracy, and reproducibility of AVN measurements in TAVR candidates by performing a comparison with MDCT.


In 150 patients with severe, symptomatic aortic stenosis referred for TAVR, data on aortic annular and root dimensions prospectively acquired using 3D TEE and MDCT were retrospectively analyzed. Image quality on 3D TEE and the duration of analysis with AVN were recorded, as well as the aortic valve Agatston score on MDCT.


Data were obtained using 3D TEE and MDCT in 100% of patients for aortic annular dimensions and in 89% for aortic root dimensions. The mean duration of analysis using AVN was 4.2 ± 1.0 min, but it was significantly shorter with better 3D echocardiographic image quality and lower Agatston score on MDCT. Correlation of measurements between 3D TEE and MDCT was good to excellent for all anatomic locations (sinotubular junction mean diameter, R = 0.71; sinus of Valsalva mean diameter, R = 0.87; aortic annular mean diameter, R = 0.75; aortic annular perimeter, R = 0.83; aortic annular area, R = 0.91), with low inter- and intraobserver variability (intraclass correlation coefficient ≥ 0.93 and r ≥ 0.90 for all locations). Comparison based on conventional prosthesis sizing charts yielded excellent agreement in prosthesis size choice (κ = 0.90).


New automated 3D transesophageal echocardiographic software allows accurate modeling and reproducible quantification of aortic annular and root dimensions with high feasibility. An excellent correlation between measurements with AVN and MDCT and agreement in prosthesis sizing suggests the use of AVN in clinical practice as potential alternative to MDCT before TAVR.


  • New automated 3D echocardiographic software (AVN) yields accurate measures of the aortic annulus.

  • AVN showed excellent agreement with MDCT in choosing TAVR prosthesis size.

  • AVN may be used for aortic root sizing in TAVR patients as an alternative to MDCT.

In patients with symptomatic severe aortic stenosis (AS) who are deemed ineligible or at high risk for surgery, transcatheter aortic valve replacement (TAVR) has rapidly become the current treatment of choice. Early TAVR experience has demonstrated the prognostic impact of complications due to inaccurate prosthesis sizing, such as paravalvular regurgitation and, less frequently, annular rupture or valve embolization. Therefore, preimplantation prosthesis sizing is considered key for procedural success and will gain even further importance as indications for TAVR expand in the near future toward an intermediate-risk population. Currently, multi–detector row computed tomography (MDCT) is the three-dimensional (3D) imaging technique of choice for aortic annular measurement and therefore for prosthesis sizing. However, this technique has its limitations in both the elderly population because of contrast nephrotoxicity and in the younger population because of radiation risks. Three-dimensional transesophageal echocardiography (TEE) is a valuable alternative to MDCT. Data from both MDCT and 3D TEE are commonly analyzed manually and require advanced expertise to accurately measure the aortic annulus. Automated 3D echocardiographic software based on adaptive analytics algorithms may help standardize the measurement, reducing the time of analysis and maximizing accuracy and reproducibility, as has been shown for postprocessing tools for MDCT. Prototype software for automated aortic annular and root sizing (Aortic Valve Navigator [AVN]; Philips Medical Systems, Andover, MA) has been recently developed, allowing reconstruction of an aortic root model from 3D transesophageal echocardiographic images. The aims of this study were (1) to assess the feasibility and reproducibility of automated 3D transesophageal analysis of the aortic annulus and root, (2) to assess the accuracy of automated 3D transesophageal analysis by comparison with measurements obtained with MDCT, and (3) to evaluate the impact on transcatheter valve prosthesis choice using automated 3D transesophageal analysis.


Population Characteristics

We retrospectively included 150 patients with severe AS who underwent TAVR, applying the following exclusion criteria: (1) lack of or inadequate preprocedural MDCT ( n = 23), (2) lack of intraprocedural 3D TEE or imaging performed using an ultrasound system from another vendor ( n = 134), and (3) valve-in-valve procedures ( n = 18). Selection criteria for TAVR were the presence of severe AS, defined by mean aortic valve gradient ≥ 40 mm Hg or aortic valve area ≤ 1 cm 2 . Eligibility for TAVR was decided in a heart team discussion and mainly for those patients deemed too high risk or having contraindications for cardiac surgery. The need for patient written informed consent was waived by the institutional review board of the Leiden University Medical Center after approval of this retrospective analysis of clinically acquired data.

Transthoracic Echocardiography

Comprehensive two-dimensional transthoracic echocardiography was performed before TAVR using a commercially available ultrasound system (Vivid; GE Vingmed Ultrasound, Horten, Norway). Valve morphology, AS severity, stroke volume, and left ventricular (LV) function were measured according to the European Association of Cardiovascular Imaging and American Society of Echocardiography standards. Specifically, LV end-diastolic and end-systolic volumes and LV ejection fraction were calculated using the Simpson biplane method. Stroke volume was indexed to body surface area.

Three-Dimensional Transesophageal Echocardiography

Acquisition of 3D transesophageal echocardiographic images was performed intraprocedurally using a commercially available ultrasound system (iE33 and EPIQ7; Philips Medical Systems) and transesophageal probe (X7-2t). Using 3D zoom mode with adjustment of lateral and elevation width, 3D images of the whole aortic valve apparatus were obtained, including the LV outflow tract (LVOT), the aortic root, and the ascending aorta. Image quality was recorded in all patients and graded as follows: 1 = poor, 2 = fair, 3 = good, 4 = very good, and 5 = excellent.

The 3D transesophageal echocardiographic images of the aortic annulus and root were subsequently analyzed using a prototype of dedicated software that allows automated modeling of the aortic annulus and aortic root in a multistep, predefined work flow. The software provides several measurements of the aortic root (LVOT, aortic annulus, sinus of Valsalva [SV], and sinotubular junction [STJ]), which can be preselected before the analysis. The following measurements were collected for the purpose of this study: aortic annular area, aortic annular maximum and minimum diameter, aortic annular perimeter, SV mean diameter, and STJ mean diameter. The mean aortic annulus diameter was calculated as the average of the maximum and minimum diameters. Figure 1 highlights the key steps of the software work flow.

Figure 1

Overview of work flow of the 3D transesophageal echocardiographic automated software. A graphical illustration (A) and an example of a case with 3D TEE (B) demonstrate the automated orientation of different MPR planes ( green frame , long-axis view; blue frame , coronal view; red frame , short-axis view), including markers for the ascending aorta (AA) and LVOT. Automated tracings of different anatomic locations are shown in (C) (aortic annulus), (D) (SV), and (E) (STJ). (F) Obtained 3D model of the three different planes ( red , aortic annulus; yellow , SV; blue , STJ) with the possibility of overlaying a 3D model of various prostheses ( pink mesh ).

After selection of the 3D transesophageal echocardiographic Digital Imaging and Communications in Medicine file, an automated algorithm orients the multiplanar reconstruction (MPR) images as requested ( Figures 1 A and 1B). After manual frame selection (in an early systolic phase), the user can place two reference markers (“AA: Ascending Aorta” or “LVOT”) to facilitate interpretation of the images and eventually further adjust the MPRs. The software uses built-in algorithms to automatically define three virtual aortic valve annular points in each MPR, which can be confirmed by the user ( Figures 1 C–1E). The MPR in the transverse view automatically adjusts to the plane containing the three points, followed by an automated generation of the annulus contour. After approval of the annular contour, the software automatically initiates the LVOT contour at a plane parallel to the aortic annular plane and offset 5 mm into the left ventricle. The same steps are then followed to generate the SV and STJ contours in their respective planes.

If deemed necessary by the user, the software allows adjustment of the obtained contours using two different approaches ( Figure 2 ). The first approach comprises manual regional adjustment of the contour by altering the position of the automated points. The second approach comprises global contour scaling (“slider”) with a user-predetermined amount (or “offset”). Users can define this specific offset by determining the difference in measurements (or “bias”) between 3D echocardiography and MDCT in a group of their own patients. Both approaches can also be combined to further improve contour tracing. Finally, after generating an aortic root model with the four different contours (aortic annulus, LVOT, SV, and STJ), the software allows the user to select from a variety of TAVR devices, allowing a 3D overlay of the device superimposed on the generated aortic root model ( Figure 1 F). An instantaneously generated and detailed list of the performed measurements is provided by the software, which can be exported to an Excel file (Microsoft, Redmond, WA) at any moment of the analysis. Duration of the analysis (in seconds) with the novel software was recorded in all patients.

Figure 2

Regional and global adjustment of the generated contour. An example of a stenotic aortic valve with calcification of the right coronary cusp is shown ( asterisk , A ). The annular contour was automatically generated excluding the calcification (B) . Regional adjustment was performed by altering the position of one of the automated points ( B , blue arrow ) more eccentrically (C) . Global adjustment can further be performed by offsetting the global contour ( C , blue arrows ), resulting in improved contour tracing (D) .

Multi–Detector Row Computed Tomography

Aortic valve data acquisition was performed using either a 64–detector row or 320–detector row computed tomographic scanner (Aquilion 64 or Aquilion ONE, respectively; Toshiba Medical Systems, Otawara, Japan). The data acquisition protocol for the Aquilion 64 and Aquilion ONE has been previously described. Scanning was performed during midinspiratory breath hold, and the arrival of contrast media was detected using a real-time tracking technique with a threshold of +180 Hounsfield units (HU). The electrocardiogram was simultaneously recorded during image acquisition for retrospective gating. After reconstruction of the data sets, offline postprocessing was performed (Vitrea FX 6.5; Vital Images, Minnetonka, MN). From the systolic images (30%–35% of the R-R interval) using MPR planes, aortic annular dimensions (maximum annular diameter, minimum annular diameter, aortic annular perimeter, and aortic valve area) and aortic root dimensions (SV and STJ) were measured according to the standard protocol. The mean annular diameter (in millimeters) was calculated as the average of the maximum and minimum diameters. From the non-contrast-enhanced images, the aortic valve calcification (AVC) score (or Agatston score of the aortic valve) was assessed.

Statistical Analysis

All normally distributed continuous data are presented as mean ± SD unless otherwise stated. All categorical variables are presented as frequencies and percentages and were compared using the χ 2 test.

To study the factors that may influence the duration of the automated 3D transesophageal echocardiographic analysis, patients were divided according to the median value of image quality on 3D TEE or of AVC score, and the differences in duration of the automated 3D echocardiographic analysis were compared using an independent-samples t test. Comparison of measurements between MDCT and 3D TEE was performed using a paired two-sided Student’s t test. Concordance between the techniques was evaluated with plots using the Bland-Altman method. To assess the correlation between measurements from echocardiography and MDCT, Pearson correlation coefficients were calculated. The first 100 patients were used as a calibration group whereby we sought to identify the average measurement bias between 3D TEE and MDCT. The last 50 patients were used as a validation cohort whereby we sought to explore whether the use of the “slider” feature, preset with the bias obtained by the calibration group as an offset, would improve the concordance between the techniques compared with the calibration cohort. To assess the agreement between the different imaging modalities regarding the choice of prosthesis size, a contingency analysis was performed, and agreement was expressed using the κ value (good agreement, κ = 0.61–0.80; very good agreement, κ = 0.81–1.00), which was also compared between the calibration and the validation groups. A receiver operating characteristic curve for AVC score was generated using bias > 1% as a classification variable.

To test interobserver variability between both methods, 10 random studies were analyzed by two cardiologists (N.A.M and E.A.P.) who independently reviewed the 3D images and performed the automatic measurements. One observer (E.A.P.) repeated the measurements in 10 randomly selected cases at two different time points to assess intraobserver variability. Interobserver and intraobserver agreements were calculated using intraclass correlation coefficients (ICCs), with excellent agreement defined as ICC > 0.75 and strong agreement defined as ICC = 0.60 to 0.74. Readers of one modality were blinded to the results of the other modality. Statistical analyses were performed using SPSS version 22.0 (IBM, Armonk, NY), and differences were considered significant at P < .05.


Patient Population

Clinical, echocardiographic, and computed tomographic characteristics of the study population are summarized in Table 1 . The study population consisted of 76 women and 74 men with mean age of 80.7 ± 7.2 years. All patients had native severe AS (mean aortic valve area, 0.8 ± 0.3 cm 2 ; mean gradient, 43.5 ± 19.5 mm Hg). The mean logistic European System for Cardiac Operative Risk Evaluation score was 18.7 ± 13.1, and the majority of procedures were performed via transfemoral access. Forty-five patients received 23-mm prostheses, 62 patients received 26-mm prostheses, 42 patients received 29-mm prostheses, and one patient received a 31-mm prosthesis.

Table 1

Clinical, echocardiographic, and computed tomographic characteristics of the study population ( N = 150)

Characteristic Value
Age (y) 80.7 ± 7.2
BMI (kg/m 2 ) 26.7 ± 5.5
Female 50.7%
EuroSCORE 18.7 ± 13.1
2D transthoracic echocardiography
AVA (cm 2 ) 0.8 ± 0.3
Mean transaortic gradient (mm Hg) 43.5 ± 19.6
LVEF (%) 50.0 ± 11.8
LV stroke volume index (mL/m 2 /beat) 38.8 ± 11.9
AVC (HU) 3,489.4 ± 1,683.1
Transfemoral vs transapical access 68.7% vs 31.3%
Clinically implanted prosthesis Edwards SAPIEN Medtronic CoreValve
23 mm 43 2
26 mm 58 4
29 mm 33 9
31 mm 0 1
Paravalvular leak
Grade 0 36.7%
Grade 1 54.6%
Grade 2 8.0%
Grade 3 0.7%
Postdilation 12.0%

2D , Two-dimensional; AVA , aortic valve area; BMI , body mass index; EuroSCORE , European System for Cardiac Operative Risk Evaluation; LVEF , LV ejection fraction.

Data are expressed as mean ± SD or as percentages or numbers.

Directly after implantation.

Duration of Analysis with Automated 3D Transesophageal Echocardiographic Software

The mean duration of analysis of aortic annular and root dimensions using the novel software was 4.2 ± 1.0 min. This duration was affected by the AVC score (mean duration, 3.6 ± 1.0 min for AVC < 3,500 HU [the median value of the population] vs 6.5 ± 1.8 min for AVC ≥ 3,500 HU; P < .001) and by 3D echocardiographic image quality (3.7 ± 1.0 min for patients with image quality 4 or 5 vs 5.5 ± 1.9 min for those with image quality 1–3, P < .001).

Comparison of Computed Tomographic versus Automated 3D Echocardiographic Measurements

Computed tomographic and automated 3D transesophageal echocardiographic measurements of aortic annular area and aortic annular minimum and maximum diameters were performed in all 150 patients, while measurements of SV and STJ diameter were possible in 133 patients (89%; Table 2 ). Including the entire patient population, automated 3D echocardiographic measurements were significantly smaller than measurements by MDCT ( P < .001) in all anatomic locations, although the absolute differences were relatively small.

Table 2

Comparison between 3D transesophageal echocardiographic and computed tomographic measurements

Measurement n 3D TEE (95% limits of agreement) MDCT (95% limits of agreement) Δ3D-TEE vs MDCT (95% limits of agreement) R P
Aortic annular area (mm 2 ) 150 450.4 (436.8 to 464.0) 460.5 (447.2 to 473.7) −10.1 (−78.5 to 58.4) 0.91 <.001
Aortic annular maximum diameter (mm) 150 26.7 (26.3 to 27.1) 27.5 (27.1 to 30.0) −0.8 (−4.5 to 2.8) 0.75 <.001
Aortic annular minimum diameter (mm) 150 21.7 (21.4 to 22.1) 21.9 (21.6 to 22.3) −0.2 (−3.6 to 3.1) 0.71 <.001
Aortic annular mean diameter (mm) 150 24.3 (24.0 to 24.6) 24.6 (24.3 to 24.9) −0.3 (−2.7 to 2.1) 0.75 <.001
Aortic annular perimeter (mm) 150 78.3 (77.3 to 79.7) 78.5 (77.1 to 79.4) −0.2 (−8.5 to 8.2) 0.83 <.001
SV mean diameter (mm) 133 33.5 (32.9 to 34.0) 34.2 (33.5 to 34.8) −0.7 (−4.2 to 2.8) 0.87 <.001
STJ mean diameter (mm) 133 27.8 (27.2 to 28.2) 28.2 (27.6 to 28.6) −0.4 (−4.8 to 4.0) 0.73 <.001

Aortic annular mean diameter = (aortic annular maximum diameter + aortic annular minimum diameter)/2.

An excellent correlation between automated 3D TEE and MDCT was observed for aortic annular area measurements ( r = 0.91). The correlations for aortic annular perimeter, annular maximum, minimum, and mean diameters, and SV and STJ mean diameters were good ( r = 0.83, r = 0.75, r = 0.71, r = 0.75, r = 0.87, and r = 0.73, respectively). Figure 3 and Supplemental Figure 1 (available at www.onlinejase.com ) show Bland-Altman plots for agreement between both methods and scatterplots to demonstrate degree of correlation. A receiver operating characteristic curve was constructed to determine a cutoff for AVC using an absolute bias > 1% as a classification variable ( Supplemental Figure 2; available at www.onlinejase.com ). Good discriminatory ability was found for MDCT-derived AVC score (area under the curve = 0.669) and yielded an AVC value of 3,366 HU as a threshold (sensitivity, 66.3%; specificity, 63.2%) in our cohort.

Figure 3

Bland-Altman plots and scatterplots of 3D transesophageal echocardiographic automated analysis versus multi–detector row computed tomographic measurements for aortic annular area and mean annular diameter and perimeter in the overall patient population. Aortic annular mean diameter = (aortic annular maximum diameter + aortic annular minimum diameter)/2.

“Slider” Feature for Global Scaling: Calibration versus Validation Cohort

After analyzing the first 100 patients (calibration cohort), an interim comparison of measurements of automated 3D TEE and MDCT showed an average bias for aortic annular area of −17.8 mm 2 (underestimation) with 95% limits of agreement of −86.1 to 50.4 mm 2 ( Figure 4 A). In the following 50 patients (validation cohort), the “slider” feature was used for automated global scaling with this bias used as an offset. Application of the “slider” resulted in improved agreement between the measurements of both modalities, with a slight overestimation of the automated 3D echocardiographic analysis software of +5.4 mm 2 and narrower 95% limits of agreement (−53.2 to 64.1 mm 2 ; Figure 4 B). Comparison of the different characteristics of the calibration and validation cohort is shown in Supplemental Table 1 (available at www.onlinejase.com ).

Apr 15, 2018 | Posted by in CARDIOLOGY | Comments Off on Feasibility, Accuracy, and Reproducibility of Aortic Annular and Root Sizing for Transcatheter Aortic Valve Replacement Using Novel Automated Three-Dimensional Echocardiographic Software: Comparison with Multi–Detector Row Computed Tomography

Full access? Get Clinical Tree

Get Clinical Tree app for offline access