Background
Intervendor consistency of left ventricular (LV) volume measurements using three-dimensional transthoracic echocardiography (3DTTE) has never been reported. Accordingly, we designed a prospective study to (1) compare head-to-head the accuracy of three three-dimensional echocardiography (3DE) systems in measuring LV volumes and ejection fraction (EF) against cardiac magnetic resonance (CMR); (2) assess the intervendor variability of LV volumes and EF; and (3) compare the accuracy of fully automated versus semiautomated (i.e., manually corrected) methods of LV endocardial delineation against CMR.
Methods
We studied 92 patients (64% males, 52 years [95% CI, 20-83]) with a wide range of end-diastolic volumes (from 87 to 446 mL) and EFs (from 16% to 77%) using three different 3DE platforms (iE33; Vivid E9; Acuson SC2000) during the same echo study. CMR was performed within 3 ± 5 hours from the 3DE study in 35 patients.
Results
LV volumes provided by the three 3DE systems correlated with CMR volumes: end-diastolic volume (iE33: R 2 = 0.93; E9: R 2 = 0.94; SC2000: R 2 = 0.94), end-systolic volume (iE33: R 2 = 0.93; E9: R 2 = 0.95; SC2000: R 2 = 0.94), and EF (iE33: R 2 = 0.79; E9: R 2 = 0.80; SC2000: R 2 = 0.77). In the 92 patients studied, LV volumes and EFs measured with the three systems were similar. Use of fully automated endocardial border detection algorithms significantly underestimated LV volumes, and the degree of underestimation was higher with larger LV volumes.
Conclusions
LV volumes and EFs measured with the three 3DE systems are consistent. Fully automated algorithms underestimated LV volumes. Our findings may help in the clinical interpretation of LV parameters obtained using different 3DE systems and encourage the clinical use of 3DTTE.
Highlights
- •
Semiautomated LV endocardial delineation provided close 3DE and CMR LV volumes.
- •
LV volumes measured with the three 3DE systems tested in this study were similar.
- •
Semiautomated algorithms provided larger LV volumes than the fully automated ones.
Fully automated methods were faster and more reporducible in measuring LV volumes.
Three-dimensional transthoracic echocardiography (3DTTE) is the most accurate technique to assess left ventricular (LV) volumes and ejection fraction (EF) using cardiac ultrasounds. All LV analysis algorithms used in commercially available three-dimensional echocardiography (3DE) systems have been independently validated against cardiac magnetic resonance (CMR), and close correlations between 3DTTE and CMR measurements have been reported, albeit with a small underestimation of LV volumes by 3DTTE. Although this assumption has never been tested, it has been assumed that all 3DE systems are equally accurate and provide consistent measurements of LV volumes and EF. Recently, a significant intervendor variability of LV three-dimensional (3D) strain was reported by our group and by other investigators, which casts some doubts on the actual consistency of LV measurements obtained with different 3DE systems.
The assessment of intervendor consistency of LV volume measurements by 3DTTE has important scientific and clinical implications. From the scientific point of view, it would be important to know whether the reference values of LV volumes and EF that have been obtained using one vendor-specific 3DE system can be applied to studies employing different 3DE systems. From the clinical point of view, physicians need to know how to deal with patients who require follow-up echocardiographic examinations for LV size and function monitoring by 3DTTE (e.g., patients with asymptomatic severe valvular heart diseases, with previous myocardial infarction or cardiac surgery, or undergoing potentially cardiotoxic chemotherapy, etc.). Can these patients be scanned by any 3DE system in the laboratory, or is it mandatory to have them examined always with the same 3DE system in order to reliably assess changes in LV volumes and EF?
To address these issues, we designed a prospective observational study to compare LV measurements obtained with three different 3DE systems from the same patient during the same study. Specifically, our aims were to (1) test the accuracy of LV volumes and EF measured with different 3DE systems in the same patient; (2) assess the intervendor variability of LV volume and EF measurements; and (3) compare the accuracy of a fully automated method (i.e., no manual editing from the operator) versus semiautomated methods (i.e., with manual editing from the operator) for LV endocardial border delineation by 3DE against CMR.
Methods
Study Population
From July to October 2015, 92 patients referred for clinically indicated echocardiography were prospectively recruited in a single tertiary center. Patients were enrolled if they were >17 years old and if the standard echocardiographic study revealed a good image quality from the apical approach (less than three segments not adequately visualized without contrast enhancement in four- and two-chamber and long-axis apical views). Exclusion criteria included pregnancy, arrhythmias, unstable clinical condition, indication for an urgent echocardiographic study, and unwillingness to participate to the study.
The size of the study population was calculated according to the results of the comparison of LV volumes measured with the three tested 3DE systems in the first 19 patients (pilot study). Accordingly, we assumed a standard deviation of the LV end-diastolic volume (EDV) of 30 mL with an average difference between EDVs measured with iE33 and Vivid E9 of 20 mL. In order to have 90% power at a two-sided type I error rate of 1%, at least 65 patients needed to be studied.
Additionally, we included 35 hemodynamically stable patients undergoing a CMR study for clinical reasons, scheduled within 24 hours from the 3DTTE study, to compare the relative accuracy in LV quantification of the three 3DE systems.
Blood pressure was measured in all subjects immediately before the echocardiographic examination. Height and weight were measured using calibrated stadiometer and scale, and body surface area was calculated according to the formulas by Dubois and Dubois. The study was approved by the Ethics Committee of the University of Padua (protocol no. 3017P/13, September 23, 2013), and signed informed consent was obtained from all patients before the 3DTTE echo study.
Three-Dimensional Transthoracic Echocardiography
At the end of the conventional two-dimensional transthoracic echocardiographic exam (2DTTE), patients fulfilling the enrollment criteria were invited to participate in the study. In these patients, dedicated 2DTTE four- and two-chamber apical (three cardiac cycles) views for LV volume calculations using biplane discs’ summation algorithm and three full-volume 3DTTE LV data sets were additionally obtained from the apical approach, using three different 3DE systems, equipped with the latest technology commercially available at the time: iE33 (Philips Medical Systems, Andover, MA) equipped with X5-1 matrix probe; Acuson SC2000 (Siemens, Mountanview, CA) equipped with the 4Z1c matrix probe, and Vivid E9 (GE Vingmed, Horten, Norway) equipped with the 4V matrix probe.
Data Set Acquisition
All patients were examined on a cutout bed, in the left lateral position, with the adjustment of image contrast, frequency, depth, and sector size to maximize frame rates and optimize LV border visualization. Care was taken to encompass the entire LV cavity within the data set, and respiratory maneuvers were applied for optimizing the data set quality when necessary. In each patient, the 2DTTE views and the three full-volume LV data sets were acquired by a single researcher having at least 3 years of experience with 3DTTE acquisitions (D.M., L.P.B., U.C.), who used the echocardiographic systems in random order according to their availability.
Full-volume LV data sets were acquired using the option providing maximum spatial and temporal resolution for each 3DE system—multibeat electrocardiogram-gated acquisition mode obtained by stitching together six consecutive subvolumes: iE33 and Vivid E9; single-beat LV full volumes: Acuson SC2000. Multislice display of 3D LV data sets was used to check that all LV segments were included and to exclude any stitching artifacts between subvolumes in multibeat acquisitions. In addition, since the expected different temporal resolutions of the data sets acquired with the three echocardiographic systems could have affected the final results, 30 randomly selected patients were reexamined by acquiring single-beat full-volume data sets (SC2000 does not allow the multibeat acquisition mode) in order to achieve similar volume rates. Data sets were digitally stored and exported to dedicated workstations for subsequent analysis.
Three-Dimensional Transthoracic Echocardiography Data Set Analysis
LV data sets were analyzed with the respective vendor-specific software package as follows: Vivid E9 data sets were analyzed using 4D Auto LVQ package (EchoPac BT 12, GE Vingmed), iE33 data sets were analyzed using 3DQ ADV (Qlab 9.0, Philips Medical Systems), and Acuson SC2000 data sets were analyzed using eSie Volume Cardiac Analysis Package-Volume Left Ventricular Analysis (eSie LVA, Siemens Healthineers).
To ensure consistent and blind measurements with all software packages, digitally stored LV data sets were analyzed by a single investigator (L.P.B.), who quantified the data sets using an approach based on equipment, not on patient. Accordingly, all LV data sets acquired with iE33 were analyzed in row using 3DQ ADV. Three days later, the same investigator quantified all data sets acquired with Vivid E9 using 4D Auto LVQ, without knowledge of the results obtained during the previous quantitative analysis. Finally, after an additional 3 days, all data sets acquired with Acuson SC2000 were analyzed with eSie LVA using the same methodology. In each session of LV quantification by 3DTTE, the data sets were analyzed in random order. The methodology of LV volume and EF analysis using the three vendor-specific software algorithms ( Figures 1-3 ) has previously been published.
Intraobserver reproducibility was assessed in 28 randomly chosen patients by the same observer (L.P.B) performing a second analysis of LV data sets acquired with the three 3DE systems tested in our study 2 months after the first analyses. Interobserver reproducibility has been tested by comparing the results of the quantitative analysis performed by L.P.B. with those performed by three investigators with specific expertise with the three 3DE software packages (D.M. for 4D Auto LVQ, R.M.L. for 3dQ ADV, and M.V. for eSie LVA) who reanalyzed the same 28 data sets acquired with the corresponding echo system.
Cardiac Magnetic Resonance
CMR was performed less than 24 hours apart from the echocardiographic examination with a 1.5-T magnet (Magnetom Avanto, Siemens Medical Systems, Erlangen, Germany) using a 12-channel phased-array coil. Short-axis cines extending from the mitral valve plane to just below the LV apex were acquired using a segmented balanced steady-state free precession sequence, effective temporal resolution of 20-25 frames per cardiac cycle and 8-mm short-axis slices with 20% interslice distance. Images were analyzed using Argus software (Siemens Medical Solutions), following current guidelines, by a single experienced investigator (A.C.) blinded to the results of the 3DE analysis. End-diastolic and end-systolic frames were defined as the frames in which the cavity sizes were largest and smallest by retrospective image review. Short-axis slices with at least 50% of the LV circumference surrounded by myocardium were included in the volume. The basal and apical slices were confirmed on long-axis views. The LV cavity was traced both in end diastole and end systole with inclusion of the papillary muscle and trabeculae in the volume. EDV and end-systolic (ESV) volumes were calculated using the discs’ summation method.
Statistical Analysis
Normal distribution of variables was assessed using the Kolmogorov-Smirnov test. Accordingly, continuous variables have been summarized as mean ± SD or as median (95% CI), while scalar variables have been reported as percentages. Differences among values obtained with the three software packages have been assessed using the analysis of variance test with Bonferroni’s post hoc analysis for normally distributed variables or the Kruskal-Wallis test otherwise. Pearson’s correlation and Bland-Altman analysis were used to assess the consistency of LV measurements obtained by the three 3DE systems and by the fully automated versus semiautomated methods for LV endocardial border delineation, as well as to analyze the accuracy of LV measurements obtained by 3DTTE versus CMR.
We also tested whether the consistency of LV volume measurements among the three 3DE systems in patients with severely dilated and not severely dilated ventricles, as well as in those with dysfunctioning and not dysfunctioning ventricles. Accordingly, we compared LV volumes and EFs among patients with EDV > 200 mL and among those with EDV ≤ 200 mL, as well as among patients with EF < 40% and among those with EF ≥ 40%.
We evaluated intra- and interobserver variability of measurements using Bland-Altman plots.
All analyses have been performed using SPSS 21.0 (SPSS, Chicago, IL) and MedCalc statistical software version 10.0.1.0. (MedCalc, Mariakerke, Belgium). Differences among variables were considered significant at P < .05.
Results
Study Population
A total of 126 consecutive patients were screened for inclusion in the study. Three patients (7%) refused to participate, and 31 patients were excluded due to poor apical acoustic window or rib artifacts ( n = 10), arrhythmias ( n = 6), temporary unavailability of one of the three 3DE scanners ( n = 8), and inadvertent cancellation of data sets from the local archive of iE33 scanner before being transferred to the dedicated workstation ( n = 7).
LV quantification using the three vendor-specific software packages was successfully performed in all remaining 92 patients. The time required to analyze the three-dimensional data sets and obtain the final results was significantly longer with 3DQ ADV and shorter with eSie LVA (5 min 58 sec ± 36 sec, 3 min 44 sec ± 29 sec, and 2 min 15 sec ± 24 sec for 3DQ ADV, Auto LVQ, and eSie LVA, respectively; P = .0001). The differences in the time required to analyze the 3D data sets were related to the level of manual interaction required for the observer to obtain the final mesh: five point clicks to initialize the software package and cumbersome editing of the initial, semiautomatic, endocardial border trace for the 3DQ ADV; two point clicks and easier manual editing of the semiautomatic endocardial border delineation by clicking single “attractors,” and automatic update of the mesh by Auto LVQ; and fully automated identification of the endocardial border with easy editing by clicking and dragging the automatic endocardial border delineation by eSie.
In the 35 patients who underwent CMR, 3DTTE data sets were collected at 3 ± 5 hour interval from the CMR study. Demographic and clinical characteristics of the study patients are summarized in Table 1 . Patients were predominantly males displaying a wide range of LV EDV (from 87 to 446 mL), ESV (from 25 to 362 mL), and EF (from 16% to 77%). Patients who underwent CMR and the overall study population showed similar LV volumes and EF ( Table 1 ).
| Study patients cohort ( N = 92) | CMR cohort ( n = 35) | P value | |
|---|---|---|---|
| Age, years | 52 (20, 83) | 44 (18, 64) | <.03 |
| Caucasian (%) | 64 (70) | 31 (89) | NS |
| Men (%) | 63 (64) | 22 (63) | NS |
| Body surface area, m 2 | 1.93 (1.58, 2.25) | 1.87 (1.22, 2.32) | NS |
| Heart rate, bpm | 68 (51, 95) | 65 (49, 93) | NS |
| Systolic blood pressure, mm Hg | 120 (93, 162) | 120 (92, 144) | NS |
| Diastolic blood pressure, mm Hg | 75 (58, 86) | 79 (49, 84) | NS |
| Clinical indication for echo (%) | NS | ||
| Cardiomyopathy | 30 (34) | 12 (34) | |
| Coronary artery disease | 29 (30) | 12 (34) | |
| Heart valve disease | 15 (16) | 7 (20) | |
| Hypertension | 6 (7) | — | |
| Congenital heart diseases | 6 (7) | 3 (9) | |
| Other | 6 (7) | 1 (3) | |
| LV EDV, mL | 165 (94, 284) ∗ | 151 (102, 293) | NS |
| LV ESV, mL | 67 (28, 198) ∗ | 66 (27, 241) | NS |
| LV EF, % | 57 (25, 74) | 57 (24, 76) | NS |
| LV stroke volume, mL | 83 (44, 138) | 89 (51, 127) | NS |
Measurement Reproducibility
Intraobserver reproducibility (bias ± limits of agreement [LOA]) of LV EDV and ESV obtained using the three vendor-specific 3DE equipment with the semiautomated option for LV analysis (i.e., manual editing of endocardial contours by the user) was 2 ± 13 and 1 ± 8 mL for 3DQ ADV; 2 ± 8 and 2 ± 6 mL for 4D Auto LVQ; and 3 ± 15 and 2 ± 8 mL for eSie LVA, respectively. Interobserver reproducibility, tested head-to-head versus three investigators with specific expertise in each of the three LV analysis software packages, has been detailed in Table 2 . Results show that there was no significant bias for any of the tested software packages.
| iE33 semiautomatic | Vivid E9 semiautomatic | Vivid E9 automatic | SC2000 semiautomatic | SC2000 automatic | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Bias | LOA | Bias | LOA | Bias | LOA | Bias | LOA | Bias | LOA | |
| EDV, mL | 6 | ±25 | 0 | ±25 | 1 | ±11 | 6 | ±30 | 1 | ±16 |
| ESV, mL | 3 | ±17 | 2 | ±19 | 0 | ±9 | 3 | ±24 | 0 | ±13 |
| EF, % | 0 | ±10 | 1 | ±10 | 0 | ±4 | 3 | ±10 | 0 | ±4 |
Intervendor Accuracy and Consistency
In the 92 study patients examined with the three 3DE systems, LV data sets were acquired at similar heart rates ( Table 3 ). Data sets acquired with Vivid E9 had a significantly higher temporal resolution than those acquired with iE33 ( Table 3 ). As expected from single-beat acquisitions, the temporal resolution of the LV data sets acquired with SC2000 was significantly lower than the temporal resolution of multibeat LV data sets acquired with either Vivid E9 or iE33 ( Table 3 ).
| iE33 semiautomatic | Vivid E9 semiautomatic | SC2000 semiautomatic | P value | |
|---|---|---|---|---|
| Heart rate, bpm | 64 (48, 95) | 65 (45, 99) | 66 (50,101) | .646 |
| Volume rate, Hz | 28 (21, 33) ∗ † | 44 (26, 61) ∗ ‡ | 21 (16, 28) † ‡ | <.0001 |
| EDV, mL | 162 (99, 290) | 165 (94, 285) | 165 (94, 277) | .967 |
| ESV, mL | 70 (28, 205) | 65 (28, 196) | 78 (30, 194) | .729 |
| EF, % | 55 (21, 76) | 57 (25, 74) | 54 (21, 70) | .514 |
| Stroke volume, mL | 82 (40, 138) | 83 (47, 138) | 76 (37, 134) | .159 |
‡ P < .05, Vivid E9 vs SC2000.
Using the semiautomated method for LV endocardial border delineation, all three 3DE systems provided LV volumes and EFs with excellent correlations with the LV measurements obtained by CMR ( Figures 4-6 ). Bland-Altman plots showed reasonable limits of agreements for LV volumes and EF between 3DE and CMR ( Figures 4-6 ). Bias was negative and of similar amount for both EDV and ESV. For all tested 3DE systems, LV volumes measured by 3DE were significantly smaller than those measured by CMR ( Table 4 ), and the slopes of the regression lines were <1 ( Figures 4-6 ).
| CMR | iE33 semiautomatic | Vivid E9 semiautomatic | SC2000 semiautomatic | |
|---|---|---|---|---|
| EDV, mL | 151 (102, 293) | 146 (92, 291) ∗ | 140 (88, 274) ∗ | 145 (89, 270) ∗ |
| ESV, mL | 66 (27, 241) | 59 (25, 179) † | 57 (25, 181) ∗ | 60 (26, 177) † |
| EF, % | 57 (24, 76) | 60 (22, 79) ‡ | 61 (24, 76) ‡ | 59 (24, 75) ‡ |
| Stroke volume, mL | 89 (51, 127) | 82 (42, 142) ‡ | 82 (44, 142) ‡ | 82 (47, 151) ‡ |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
