Advertisement

Left ventricular volume and ejection fraction measurements by fully automated 3D echocardiography left chamber quantification software versus CMR: A systematic review and meta-analysis

Published:September 01, 2022DOI:https://doi.org/10.1016/j.jjcc.2022.08.007

      Highlights

      • Adoption of three-dimensional echocardiography (3DE) has been limited due to time constraints and required expertise.
      • Automated quantification has aimed at enhancing both accuracy and efficiency.
      • Compared to cardiac magnetic resonance (CMR), 3DE underestimates left ventricular volumes with differences <20 mL.
      • Compared to CMR, left ventricular ejection fraction showed similar values.
      • Fully automated 3DE left chamber quantification makes routine examinations practical.

      Abstract

      Background

      Although transthoracic three-dimensional echocardiography (3DE) is now recommended by guidelines for left ventricular (LV) volumetric measurements, widespread implementation has been limited due to time constraints and required expertise. We hypothesized that fully automated 3DE left chamber quantification software might provide accurate measurements, and that its application could eliminate these obstacles.

      Methods

      To address this hypothesis, we conducted a systematic review and meta-analysis following a search for studies that compared LV volumes and ejection fraction (EF) using fully automated 3DE software (HeartModel or Dynamic HeartModel, Philips Healthcare, Andover, MA, USA) with cardiac magnetic resonance (CMR), from 2015 to 2021. A random effects model was used to determine biases, correlations, and 95 % confidence intervals (CI) of LV end-diastolic volume (EDV), end-systolic volume (ESV), and EF. Subgroup and meta-regression analyses were performed to determine effects of moderators on the outcome.

      Results

      Of 12 studies (616 subjects), mean differences and 95 % CIs in EDV, ESV, and EF between fully automated 3DE software and CMR were −19.6 mL (95 % CI; −27.6 to −11.5 mL), −11.4 mL (−16.7 to −6.2 mL), and 0.4 % (−1.1 to 2.0 %), respectively. Corresponding correlation values between the two methods were 0.91 (0.86–0.94), 0.89 (0.82–0.93), and 0.85 (0.81–0.88), respectively. Meta-regression analysis revealed that there were no effects of either publication year, type of software, or type of analysis on the outcome of LV volumetric and functional parameters except for publication year on LVESV correlation values.

      Conclusions

      Although 3DE still underestimates LV volumes, the observed differences were no >20 mL. EF showed similar values to CMR. Excellent correlations between the two techniques make fully automated 3DE left chamber quantification software useful for routine clinical practice in adult population.

      Graphical abstract

      Keywords

      Introduction

      Three-dimensional echocardiography (3DE) is an accurate method to quantify heart chamber volumes and functioning in various cardiovascular diseases [
      • Lang R.M.
      • Addetia K.
      • Narang A.
      • Mor-Avi V.
      3-dimensional echocardiography latest developments and future directions.
      ]. Current advancements in 3DE quantification software have been aimed mainly at development of fully automated analysis without manual input, in order to eliminate measurement variability. Incorporation of artificial intelligence further enhances its accuracy, resulting in more reliable estimates of left ventricular (LV) volumes and ejection fraction (EF), despite broad variation of image quality. HeartModel (HM), and more recently, Dynamic HeartModel (DHM) (Philips Healthcare, Andover, MA, USA) is commercially available, fully automated 3DE quantification software for measurement of left chamber volumes, and EFs [
      • Medvedofsky D.
      • Mor-Avi V.
      • Amzulescu M.
      • Fernández-Golfín C.
      • Hinojar R.
      • Monaghan M.J.
      • et al.
      Three-dimensional echocardiographic quantification of the left-heart chambers using an automated adaptive analytics algorithm: multicentre validation study.
      ,
      • Barbieri A.
      • Bursi F.
      • Camaioni G.
      • Maisano A.
      • Imberti J.F.
      • Albini A.
      • et al.
      Echocardiographic left ventricular mass assessment: correlation between 2D-derived linear dimensions and 3-dimensional automated, machine learning-based methods in unselected patients.
      ]. The moving contours of LV borders and waveforms can provide a holistic view of LV volume changes during the cardiac cycle to assess diagnostic confidence.
      We previously conducted a systematic review and meta-analysis to compare LV volumes and LVEF with semi-automated or fully automated (including HM) 3DE quantification software with corresponding values derived from cardiac magnetic resonance (CMR) imaging [
      • Kitano T.
      • Nabeshima Y.
      • Otsuji Y.
      • Takeuchi M.
      Accuracy of left ventricular volumes and ejection fraction measurements by contemporary three-dimensional echocardiography with semi- and fully automated software: systematic review and meta-analysis of 1,881 subjects.
      ]. Although 3DE still underestimated LV volumes, LVEF showed similar values to CMR for both types of software [
      • Kitano T.
      • Nabeshima Y.
      • Otsuji Y.
      • Takeuchi M.
      Accuracy of left ventricular volumes and ejection fraction measurements by contemporary three-dimensional echocardiography with semi- and fully automated software: systematic review and meta-analysis of 1,881 subjects.
      ,
      • Généreux P.
      • Stone G.W.
      • O'Gara P.T.
      • Marquis-Gravel G.
      • Redfors B.
      • Giustino G.
      • et al.
      Natural history, diagnostic approaches, and therapeutic strategies for patients with asymptomatic severe aortic stenosis.
      ]. However, as with other meta-analysis publications, we only analyzed mean differences in LV volumes and LVEF between the two imaging modalities. Assessment of correlation between the two techniques is also important. Since then, there have been several studies that compared LV volumetric quantification by 3DE using HM or DHM with CMR. Therefore, in this study, we conducted a systemic review and meta-analysis of comparisons of LV volumes and LVEF between fully automated 3DE quantification software and CMR, focusing on mean differences and correlations.

      Methods

      Search strategy

      We followed PRISMA (Preferred Reporting Items for Systematic reviews and Meta-Analysis) guidelines in conducting this systematic review and meta-analysis. Two authors (VCW and TK) used three electronic databases (PubMed, Embase, and Scopus) to systematically search for studies that described comparisons of LV volumes [LV end-diastolic volume (EDV) and LV end-systolic volume (ESV)] and LVEF using transthoracic 3DE and CMR (accessed on November 23, 2021). Key terms used were “three-dimensional echocardiography”, “3D echocardiography”, “magnetic resonance imaging”, “cardiovascular magnetic resonance”, “cardiac magnetic resonance”, “left ventricular”, and “left ventricle”. We selected only human, adult, and full-text studies. Search strategies used are listed in Online Table 1. Since the first study using fully automated 3DE left chamber quantification software was published in 2015, we searched for articles that investigated use of fully automated software from January 1, 2015, to November 23, 2021.

      Study selection

      From the search results, we included studies if they described results of comparisons of LVEDV, LVESV, and LVEF using 3DE and CMR in adult patients. We included only studies that used fully automated 3DE quantification software that was either HM or DHM. In addition, both fully automated analysis and manually edited analysis, when performed, were included in our analysis. Fully automated analysis was defined as a software algorithm that automatically generated LV casts, followed by LV volume determination without manual input. Manual editing was defined as manual tracing performed after the fully automated analysis because the examiner felt that border delineation was inaccurate in some or all of the LV endocardial border in 3D space.

      Data extraction

      Quantitative data collected included means ± standard deviations (SDs) and correlation coefficients of LVEDV, LVESV, and LVEF using both 3DE and CMR on the same subjects in each study. In addition to fundamental parameters, such as age and gender, we collected information regarding the method of patient selection, volume rate of 3DE datasets, type of software used for the analysis (HM or DHM), type of analysis (fully automated or manual editing in addition to fully automated quantification), and the LV border threshold value used for the analysis.

      Statistical analysis

      Continuous data were expressed as means ± SDs or medians and interquartile ranges, according to the data distribution. Categorical variables were presented as absolute numbers or percentages. Feasibility, mean difference, and correlation were assessed by ‘metaprop’, ‘metacont’ and ‘metacor’ in package of “meta.” Mean differences or correlations and their 95 % confidence intervals (CIs) between 3DE and CMR were computed using random effects models and were graphically presented as forest plots. Funnel plots were constructed, and the Egger test was used to assess publication bias [
      • Egger M.
      • Davey Smith G.
      • Schneider M.
      • Minder C.
      Bias in meta-analysis detected by a simple, graphical test.
      ]. Heterogeneity between subgroups was assessed using the Cochran Q test and the inconsistency factor (I2). Outlier, influence, and leave-one-out analyses were also performed to detect outliers/influential cases. Meta-regression analyses were performed to examine possible factors for the outcome using the “metafor” package. Quality of studies was assessed using the quality assessment tools of Downs and Black [
      • Downs S.H.
      • Black N.
      The feasibility of creating a checklist for the assessment of the methodological quality both of randomised and nonrandomised studies of health care interventions.
      ]. A two-sided p-value <0.05 was considered statistically significant except in the Egger test, where p < 0.1 was used. All statistical analyses were performed using commercial software (R version 4.0.5, The R Foundation for Statistical Computing, Vienna, Austria).

      Results

      Fig. 1 shows the PRISMA flow chart illustrating the selection process used in this study. Among 5760 titles searched from three databases, we excluded 4971 titles due to duplication, non-human, non-adult, non-full-text article, or publication before 2015. We further excluded 775 articles with inappropriate titles or abstracts, leaving 13 articles for further full-text assessment. We excluded one article that described only left atrial volumes. After reading the full text, 12 articles with 616 subjects were selected for analysis [
      • Tsang W.
      • Salgo I.S.
      • Medvedofsky D.
      • Takeuchi M.
      • Prater D.
      • Weinert L.
      • et al.
      Transthoracic 3D echocardiographic left heart chamber quantification using an automated adaptive analytics algorithm.
      ,
      • Yang L.T.
      • Nagata Y.
      • Otani K.
      • Kado Y.
      • Otsuji Y.
      • Takeuchi M.
      Feasibility of one-beat real-time full-volume three-dimensional echocardiography for assessing left ventricular volumes and deformation parameters.
      ,
      • Levy F.
      • Schouver E.D.
      • Iacuzio L.
      • Civaia F.
      • Rusek S.
      • Dommerc C.
      • et al.
      Performance of new automated transthoracic three-dimensional echocardiographic software for left ventricular volumes and function assessment in routine clinical practice: comparison with 3 tesla cardiac magnetic resonance.
      ,
      • Tamborini G.
      • Piazzese C.
      • Lang R.M.
      • Muratori M.
      • Chiorino E.
      • Mapelli M.
      • et al.
      Feasibility and accuracy of automated software for transthoracic three-dimensional left ventricular volume and function analysis: comparisons with two-dimensional echocardiography, three-dimensional transthoracic manual method, and cardiac magnetic resonance imaging.
      ,
      • Żygadło A.
      • Kaźnica-Wiatr M.
      • Błaut-Jurkowska J.
      • Knap K.
      • Lenart-Migdalska A.
      • Smaś-Suska M.
      • et al.
      Evaluation of the clinical suitability of automated left ventricle's fraction and volume measurements in 3-dimensional echocardiography compared to values obtained in magnetic resonance imaging (pilot study).
      ,
      • Barletta V.
      • Hinojar R.
      • Carbonell A.
      • González-Gómez A.
      • Fabiani I.
      • Di Bello V.
      • et al.
      Three-dimensional full automated software in the evaluation of the left ventricle function: from theory to clinical practice.
      ,
      • Levy F.
      • Marechaux S.
      • Iacuzio L.
      • Schouver E.D.
      • Castel A.L.
      • Toledano M.
      • et al.
      Quantitative assessment of primary mitral regurgitation using left ventricular volumes obtained with new automated three-dimensional transthoracic echocardiographic software: a comparison with 3-tesla cardiac magnetic resonance.
      ,
      • Narang A.
      • Mor-Avi V.
      • Prado A.
      • Volpato V.
      • Prater D.
      • Tamborini G.
      • et al.
      Machine learning based automated dynamic quantification of left heart chamber volumes.
      ,
      • Volpato V.
      • Mor-Avi V.
      • Narang A.
      • Prater D.
      • Gonçalves A.
      • Tamborini G.
      • et al.
      Automated, machine learning-based, 3D echocardiographic quantification of left ventricular mass.
      ,
      • Wu V.C.
      • Kitano T.
      • Nabeshima Y.
      • Otani Y.
      • Chu P.H.
      • Takeuchi M.
      Optimal threshold of three-dimensional echocardiographic fully automated software for quantification of left ventricular volumes and ejection fraction: comparison with cardiac magnetic resonance disk-area summation method and feature tracking method.
      ,
      • Levy F.
      • Iacuzio L.
      • Schouver E.D.
      • Essayagh B.
      • Civaia F.
      • Dommerc C.
      • et al.
      Performance of a new fully automated transthoracic three-dimensional echocardiographic software for quantification of left cardiac chamber size and function: comparison with 3 tesla cardiac magnetic resonance.
      ,
      • Italiano G.
      • Tamborini G.
      • Fusini L.
      • Mantegazza V.
      • Doldi M.
      • Celeste F.
      • et al.
      Feasibility and accuracy of the automated software for dynamic quantification of left ventricular and atrial volumes and function in a large unselected population.
      ]. In the 12 studies, patients were included if they were in sinus rhythm or atrial fibrillation, and agreed to participate; therefore these patients included ischemic heart disease, valvular heart disease, dilated cardiomyopathy, pericardial disease, hypertrophic cardiomyopathy, and secondary cardiomyopathy. Two studies included congenital heart disease while others did not include or did not specify. Patients were excluded if they had poor endocardial visualization on 2D echocardiography of ≥3 contiguous segments using a 17-segment model. Table 1 depicts the characteristics of included studies. There were 7 studies (n = 323) that used HM and 5 studies (n = 293) that used DHM. Among these studies, all described LV volumes and LVEF in terms of means ± SDs. Manuscript quality is shown in Online Table 2.

      Image quality and feasibility

      There were 5 studies described image quality of the echocardiography exams, and all these studies reported number of patients with poor image quality (Online Table 3). Pooled prevalence of poor image quality was 14.7 % (95 % CI: 7.1 % to 28.2 %). All but one of the 12 studies reported feasibility of 3DE analysis. Combined feasibility was 95.8 % (95 % CI: 88.1 to 98.6 %).

      3DE and CMR measurements

      Fig. 2A, B, and C presents forest plots of the mean differences in LVEDV, LVESV, and LVEF between the two methods. Fully automated 3DE software significantly underestimated LVEDV (mean differences: −19.6 mL, 95 % CI: −27.6 to −11.5 mL, p < 0.001), and LVESV (−11.4 mL, −16.7 to −6.2 mL, p < 0.001) versus CMR. However, there were no significant differences in LVEF (0.4 %, −1.1 % to 2.0 %, p = 0.557). Fig. 3A, B, and C present corresponding correlation values between the two techniques. Correlation values were excellent with moderate to severe between study heterogeneity (I2: 84 %, 78 %, and 40 % for LVEDV, LVESV, and LVEF, respectively).
      Fig. 1
      Fig. 1PRISMA flow chart of study selection. The systematic review and meta-analysis were conducted according to PRISMA guidelines.
      Table 1Fundamental characteristics in 12 studies (n = 616).
      First authorYearNumber (men)AgeVolume rateMachineProbeSoftwareAuto/manualThresholdDisease type
      Tsang
      • Tsang W.
      • Salgo I.S.
      • Medvedofsky D.
      • Takeuchi M.
      • Prater D.
      • Weinert L.
      • et al.
      Transthoracic 3D echocardiographic left heart chamber quantification using an automated adaptive analytics algorithm.
      201665 (31)50 ± 1716.0EPIQ/iE33X5HMFully automatedNADiverse
      Yang
      • Yang L.T.
      • Nagata Y.
      • Otani K.
      • Kado Y.
      • Otsuji Y.
      • Takeuchi M.
      Feasibility of one-beat real-time full-volume three-dimensional echocardiography for assessing left ventricular volumes and deformation parameters.
      201634 (27)64 ± 1219.6EPIQX5-1HMManually editedNADiverse
      Levy
      • Levy F.
      • Schouver E.D.
      • Iacuzio L.
      • Civaia F.
      • Rusek S.
      • Dommerc C.
      • et al.
      Performance of new automated transthoracic three-dimensional echocardiographic software for left ventricular volumes and function assessment in routine clinical practice: comparison with 3 tesla cardiac magnetic resonance.
      201754 (40)63 ± 1319.6EPIQX5-1HMManually edited80/40Diverse
      Tamborini
      • Tamborini G.
      • Piazzese C.
      • Lang R.M.
      • Muratori M.
      • Chiorino E.
      • Mapelli M.
      • et al.
      Feasibility and accuracy of automated software for transthoracic three-dimensional left ventricular volume and function analysis: comparisons with two-dimensional echocardiography, three-dimensional transthoracic manual method, and cardiac magnetic resonance imaging.
      201784 (NS)59 ± 15NSEPIQ/iE33X5HMFully automated50/50Diverse
      Zygadlo
      • Żygadło A.
      • Kaźnica-Wiatr M.
      • Błaut-Jurkowska J.
      • Knap K.
      • Lenart-Migdalska A.
      • Smaś-Suska M.
      • et al.
      Evaluation of the clinical suitability of automated left ventricle's fraction and volume measurements in 3-dimensional echocardiography compared to values obtained in magnetic resonance imaging (pilot study).
      201713 (7)51 ± 19NSEPIQX5-1HMFully automatedNADiverse
      Barletta
      • Barletta V.
      • Hinojar R.
      • Carbonell A.
      • González-Gómez A.
      • Fabiani I.
      • Di Bello V.
      • et al.
      Three-dimensional full automated software in the evaluation of the left ventricle function: from theory to clinical practice.
      201820 (NS)60 ± 18NSEPIQX5-1HMFully automated74/68Diverse
      Levy
      • Levy F.
      • Marechaux S.
      • Iacuzio L.
      • Schouver E.D.
      • Castel A.L.
      • Toledano M.
      • et al.
      Quantitative assessment of primary mitral regurgitation using left ventricular volumes obtained with new automated three-dimensional transthoracic echocardiographic software: a comparison with 3-tesla cardiac magnetic resonance.
      201853 (37)64 ± 1220.0EPIQX5-1HMManually edited80/40MR
      Narang
      • Narang A.
      • Mor-Avi V.
      • Prado A.
      • Volpato V.
      • Prater D.
      • Tamborini G.
      • et al.
      Machine learning based automated dynamic quantification of left heart chamber volumes.
      201920 (8)54 ± 1919.0EPIQX5-1DHMManually editedNADiverse
      Volpano
      • Volpato V.
      • Mor-Avi V.
      • Narang A.
      • Prater D.
      • Gonçalves A.
      • Tamborini G.
      • et al.
      Automated, machine learning-based, 3D echocardiographic quantification of left ventricular mass.
      201920 (9)55 ± 1919.0EPIQX5-1DHMManually edited60/30Diverse
      Wu
      • Wu V.C.
      • Kitano T.
      • Nabeshima Y.
      • Otani Y.
      • Chu P.H.
      • Takeuchi M.
      Optimal threshold of three-dimensional echocardiographic fully automated software for quantification of left ventricular volumes and ejection fraction: comparison with cardiac magnetic resonance disk-area summation method and feature tracking method.
      201957 (39)71 (64–79)NSEPIQX5-1DHMFully automated80/80Diverse
      Levy
      • Levy F.
      • Iacuzio L.
      • Schouver E.D.
      • Essayagh B.
      • Civaia F.
      • Dommerc C.
      • et al.
      Performance of a new fully automated transthoracic three-dimensional echocardiographic software for quantification of left cardiac chamber size and function: comparison with 3 tesla cardiac magnetic resonance.
      201956 (46)63 ± 1319.9EPIQX5-1DHMFully automated80/40Diverse
      Italiano
      • Italiano G.
      • Tamborini G.
      • Fusini L.
      • Mantegazza V.
      • Doldi M.
      • Celeste F.
      • et al.
      Feasibility and accuracy of the automated software for dynamic quantification of left ventricular and atrial volumes and function in a large unselected population.
      2021140 (106)63 ± 14NSEPIQX5-1DHMManually editedNADiverse
      DHM, Dynamic HeartModel; HM, HeartModel; MR, mitral regurgitation; NA, not available.
      Fig. 2
      Fig. 2Forest plots of mean differences of LVEDV (A), LVESV (B), and LVEF (C) between 3DE and CMR, and correlation (B) in LVEDV between 3DE and CMR. 3DE, three-dimensional echocardiography; CMR, cardiac magnetic resonance; LVEDV, left ventricular end-diastolic volume; LVEF, left ventricular ejection fraction; LVESV, left ventricular end-systolic volume.
      Fig. 3
      Fig. 3Forest plots of correlation of LVEDV (A), LVESV (B), and LVEF (C) between 3DE and CMR. 3DE, three-dimensional echocardiography; CMR, cardiac magnetic resonance; LVEDV, left ventricular end-diastolic volume; LVEF, left ventricular ejection fraction; LVESV, left ventricular end-systolic volume.

      Heterogeneity analysis

      All analyses regarding mean differences of LV volumes and EF between the two techniques showed that there were at most, slight heterogeneities within the studies (I2: 12 %, 0 %, and 0 %, respectively). Egger's test did not indicate the presence of asymmetry among these analyses. For LVEDV and LVESV, there were no outliers, but influence analysis revealed that Italiano's study had the largest Cook's Distance, DFFITS value, and Hat value, and influence analysis flagged the Italiano study as an influential case [
      • Italiano G.
      • Tamborini G.
      • Fusini L.
      • Mantegazza V.
      • Doldi M.
      • Celeste F.
      • et al.
      Feasibility and accuracy of the automated software for dynamic quantification of left ventricular and atrial volumes and function in a large unselected population.
      ] (Online Fig. 1A, B). Leave-one-out analysis showed that omitting the Italiano study reduced mean differences in both LVEDV and LVESV (Online Fig. 2A, B). Regarding LVEF, there were no outliers or influential cases (Online Fig. 1C). Leave-one-out analysis showed that omitting Italiano's study reduced mean differences in LVEF, resulting in the value closest to 0 (Online Fig. 2C).
      Analyses regarding correlation of LV volumes and EF between the two techniques showed that there were substantial heterogeneities within the studies (I2: 84 %, 78 %, and 40 %). However, Egger's test did not indicate the presence of asymmetry. For LVEDV, analysis indicated that Italiano's study was an outlier. Influence analysis revealed that Italiano's study had the largest Cook's Distance, DFFITS value, and Hat value, and influence analysis function regarded Italiano's study as an influential case (Online Fig. 3A). Leave-one-out analysis showed that omitting the Italiano study increased the r-value to 0.92 with a reduction of I2 to 13.2 % (Online Fig. 4A). Regarding LVESV, the Zygadlo study was flagged as an outlier and an influential case (Online Fig. 3B). Leave-one-out analysis showed that omitting Zygaldo's study increased the r-value to 0.90, but still with a high I2 of 68.9 % (Online Fig. 4B). For LVEF, there was no outlier. Influence analysis revealed that the Levy study had the largest Cook's Distance, DFFITS value, and Hat value, and influence analysis function regarded a Levy's study as an influential case (Online Fig. 3C). Leave-one-out analysis showed that omitting Levy's study did not change the r-value, but reduced I2 to 25.5 % (Online Fig. 4C).

      Subgroup analysis

      First, we divided 12 studies into two groups according to the type of the software (e.g. HM or DMH). There were no significant mean differences in LVEDV between HM (−21.2 mL, 95 % CI: −27.8 to −14.6 mL) and DHM (−16.3 mL, 95 % CI: −37.4 to 4.7 mL, p = 0.545), nor were significant mean differences in LVESV observed between HM (−10.2 mL, 95 % CI: −16.6 to −3.8 mL) and DHM (−12.1 mL, 95 % CI: −24.4 to 0.15 mL, p = 0.710). There were no significant mean differences in LVEF between HM (−0.16 %, 95 % CI: −3.2 to 2.9 %) and DHM (1.1 %, 95 % CI: −0.5 to 2.7 %, p = 0.375).
      Second, we divided studies into two groups according to the type of analysis (fully automated or manually edited). There were no significant mean differences in LVEDV between fully automated analysis (−17.6 mL, 95 % CI: −27.7 to −7.5 mL) and manually edited analysis (−21.0 mL, 95 % CI: −36.4 to −5.7 mL, p = 0.631). There were also no significant mean differences in LVESV between fully automated analysis (−9.5 mL, 95 % CI: −14.7 to −4.2 mL) and manually edited analysis (−12.7 mL, 95 % CI: −23.1 to −2.4 mL, p = 0.466). However, there were significant mean differences in LVEF between the two methods (fully automated analysis: −1.4 %, 95 % CI: −3.8 to 1.1 % vs. manually edited approach: 1.5 %, 95 % CI: −0.5 to 3.4 %, p = 0.021).

      Meta-regression analysis

      We performed meta-regression analysis to determine whether publication year, type of software, or type of analysis influenced outcomes. None of these moderators significantly influenced either mean differences or r-values of LVEDV, LVESV, and LVEF except for publication year on r-value of LVESV (estimate: −0.015, p = 0.049).

      Discussion

      The major findings in this systematic review and meta-analysis can be summarized as follows: (1) Feasibility for fully automated 3DE left chamber quantification software was excellent; (2) The software still significantly underestimated LV volumes compared with CMR by 10 to 20 mL; (3) However, similar LVEF values were observed between the two modalities; (4) Although there was no within-study heterogeneity when analyzing mean differences, moderate to severe heterogeneity was observed when assessing correlations; (5) Influence analysis revealed that there were some influential cases; (6) Meta-regression analysis showed that there were no significant associations between type of software, or type of analysis and outcomes.

      Advantage for 3DE fully automated analysis

      Multiple studies and meta-analysis have shown that 3DE is more accurate and reproducible than 2D echocardiography when compared to CMR [
      • Shimada Y.J.
      • Shiota T.
      A meta-analysis and investigation for the source of bias of left ventricular volumes and function by three-dimensional echocardiography in comparison with magnetic resonance imaging.
      ,
      • Dorosz J.L.
      • Lezotte D.C.
      • Weitzenkamp D.A.
      • Allen L.A.
      • Salcedo E.E.
      Performance of 3-dimensional echocardiography in measuring left ventricular volumes and ejection fraction: a systematic review and meta-analysis.
      ,
      • Rigolli M.
      • Anandabaskaran S.
      • Christiansen J.P.
      • Whalley G.A.
      Bias associated with left ventricular quantification by multimodality imaging - a systematic review and meta-analysis.
      ], since foreshortened views and geometric assumptions that frequently lead to errors in volume measurement can be eliminated. However, despite the well-known advantages of 3DE, this modality is not routinely adopted in clinical situations due to (1) the additional time required for analysis and (2) expertise in 3DE needed to obtain reliable results, all of which slows its adoption in routine echocardiography laboratory workflow. Therefore, newly available fully automated quantification software such as HM and DHM, has been developed to accommodate an increased need for clinical efficiency. Fully automated 3DE left chamber quantification software uses an adaptive analytical algorithm that consists of a database of LV and left atrial (LA) casts to simultaneously detect LV and LA endocardial surfaces throughout the cardiac cycle, followed by patient-specific adaptation. Although the algorithm was designed to adjust to a variety of imaging conditions, including variations in dropout, acoustic clutter, ventricular shape, and cardiac orientation relative to the transducer, manual editing is still necessary in some cases.

      Previous studies

      To the best of our knowledge, there are four published meta-analyses regarding the measurement accuracy of 3DE for LV volumes and LVEF against CMR as a reference [
      • Kitano T.
      • Nabeshima Y.
      • Otsuji Y.
      • Takeuchi M.
      Accuracy of left ventricular volumes and ejection fraction measurements by contemporary three-dimensional echocardiography with semi- and fully automated software: systematic review and meta-analysis of 1,881 subjects.
      ,
      • Shimada Y.J.
      • Shiota T.
      A meta-analysis and investigation for the source of bias of left ventricular volumes and function by three-dimensional echocardiography in comparison with magnetic resonance imaging.
      ,
      • Dorosz J.L.
      • Lezotte D.C.
      • Weitzenkamp D.A.
      • Allen L.A.
      • Salcedo E.E.
      Performance of 3-dimensional echocardiography in measuring left ventricular volumes and ejection fraction: a systematic review and meta-analysis.
      ,
      • Rigolli M.
      • Anandabaskaran S.
      • Christiansen J.P.
      • Whalley G.A.
      Bias associated with left ventricular quantification by multimodality imaging - a systematic review and meta-analysis.
      ]. Table 2 summarizes the main findings in previous meta-analyses and in the current analysis. Shimada et al. [
      • Shimada Y.J.
      • Shiota T.
      A meta-analysis and investigation for the source of bias of left ventricular volumes and function by three-dimensional echocardiography in comparison with magnetic resonance imaging.
      ] performed a meta-analysis with 95 articles (3055 patients) published from 1996 to 2010 to assess the bias of LV volumes and LVEF measured by 3DE with manual tracing or a semiautomated method and CMR. Although 3DE significantly underestimated LVEDV (−9.9 mL) and LVESV (−4.7 mL), LVEF was similar between the two techniques. Dorosz et al. performed a meta-analysis using 23 articles (1638 patients) published from 1990 to 2011 utilizing 3DE with manual tracing or a semiautomated method [
      • Dorosz J.L.
      • Lezotte D.C.
      • Weitzenkamp D.A.
      • Allen L.A.
      • Salcedo E.E.
      Performance of 3-dimensional echocardiography in measuring left ventricular volumes and ejection fraction: a systematic review and meta-analysis.
      ]. Their results showed that biases of 3DE compared with CMR were −19.1 mL, −10.1 mL, and −0.6 % for LVEDV, LVESV, and LVEF, respectively. Rigolli et al. [
      • Rigolli M.
      • Anandabaskaran S.
      • Christiansen J.P.
      • Whalley G.A.
      Bias associated with left ventricular quantification by multimodality imaging - a systematic review and meta-analysis.
      ] performed a meta-analysis in 31 articles (1198 patients) published from 1995 to 2015. Their results showed again that even though 3DE significantly underestimated LVEDV (−14.2 mL) and LVESV (−6.5 mL), differences in LVEF values were very small (0.1 %).
      Table 2Main findings from previous meta-analysis and current study.
      First authorYearSearch periodSoftwareNo. of studyNo. of samplesBias of EDV95 % CIBias of ESV95 % CIBias of EF95 % CI
      Shimada
      • Shimada Y.J.
      • Shiota T.
      A meta-analysis and investigation for the source of bias of left ventricular volumes and function by three-dimensional echocardiography in comparison with magnetic resonance imaging.
      20111996–2010Manual or semiautomated953055−9.9−11,8, −8.0−4.7−5.6, −3.7−0.1−0.5, 0.2
      Dorosz
      • Dorosz J.L.
      • Lezotte D.C.
      • Weitzenkamp D.A.
      • Allen L.A.
      • Salcedo E.E.
      Performance of 3-dimensional echocardiography in measuring left ventricular volumes and ejection fraction: a systematic review and meta-analysis.
      20121990–2011Manual or semiautomated231638−19.1−10.1−0.6
      Rigolli
      • Rigolli M.
      • Anandabaskaran S.
      • Christiansen J.P.
      • Whalley G.A.
      Bias associated with left ventricular quantification by multimodality imaging - a systematic review and meta-analysis.
      20161995–2015Manual or semiautomated311198−14.2−18.7, −9.7−6.5−9.9, −3.10.1−0.9, 1.2
      Kitano
      • Kitano T.
      • Nabeshima Y.
      • Otsuji Y.
      • Takeuchi M.
      Accuracy of left ventricular volumes and ejection fraction measurements by contemporary three-dimensional echocardiography with semi- and fully automated software: systematic review and meta-analysis of 1,881 subjects.
      20192011–2018Semiautomated271259−39.3−49.2, −29.5−19.6−25.9, −13.3−0.6−2.0, 0.7
      Fully automated11622−14.5−25.7, −3.3−6.3−11.4, −1.2−1.1−3.5, 1.3
      Current study20212015–2021Fully automated12616−19.6−27.6, −11.5−11.4−16.7, −6.20.4−1.1, 1.2
      CI, confidence interval; EDV, end-diastolic volume; EF, ejection fraction; ESV, end-systolic volume.
      Recently Kitano et al. [
      • Kitano T.
      • Nabeshima Y.
      • Otsuji Y.
      • Takeuchi M.
      Accuracy of left ventricular volumes and ejection fraction measurements by contemporary three-dimensional echocardiography with semi- and fully automated software: systematic review and meta-analysis of 1,881 subjects.
      ] performed a meta-analysis of 27 articles conducted from 2011 to 2018, including 1259 patients to assess biases of LV volumes and EF measured by 3DE with a semiautomated method compared with CMR. The authors found that 3DE significantly underestimated LVEDV (−39.3 mL) and LVESV (−19.6 mL), but LVEF bias was not significant (−0.6 %). In addition, the authors also performed a meta-analysis of 11 articles (622 patients) to assess measurement differences between 3DE with fully automated quantification software and CMR. 3DE with fully automated analysis still significantly underestimated LVEDV (−14.5 mL) and LVESV (−6.3 mL), but the LVEF was similar. They reported that fully automated 3DE quantification software seemed to have a lower bias than semiautomated 3DE software, against CMR as a reference. Although these meta-analyses showed similar results, none of these studies described the results of association of observed values between the two techniques.

      Current study

      We demonstrated that 3DE with fully automated software still significantly underestimated both LVEDV and LVESV compared with CMR, but there were no significant mean differences in LVEF, which agreed with previous studies [
      • Kitano T.
      • Nabeshima Y.
      • Otsuji Y.
      • Takeuchi M.
      Accuracy of left ventricular volumes and ejection fraction measurements by contemporary three-dimensional echocardiography with semi- and fully automated software: systematic review and meta-analysis of 1,881 subjects.
      ,
      • Shimada Y.J.
      • Shiota T.
      A meta-analysis and investigation for the source of bias of left ventricular volumes and function by three-dimensional echocardiography in comparison with magnetic resonance imaging.
      ,
      • Dorosz J.L.
      • Lezotte D.C.
      • Weitzenkamp D.A.
      • Allen L.A.
      • Salcedo E.E.
      Performance of 3-dimensional echocardiography in measuring left ventricular volumes and ejection fraction: a systematic review and meta-analysis.
      ,
      • Rigolli M.
      • Anandabaskaran S.
      • Christiansen J.P.
      • Whalley G.A.
      Bias associated with left ventricular quantification by multimodality imaging - a systematic review and meta-analysis.
      ]. Although I2 values showed no more than slight between-study heterogeneities, influence analysis revealed that there was one potentially influential case regarding LV volumes. When we removed this case, underestimation of LV volumes further diminished. In subgroup analysis, we did not find any differences in bias between the two software packages. However, bias of LVEF was significantly different between fully automated analysis and manually edited analysis. Meta-regression analysis showed that none of the moderators, including publication year, type of software, or type of analysis had a significant association with mean differences in LV volumes and LVEF. These results suggest that other factors, like spatial resolution, could be the cause of 3DE's underestimation of LV volumes compared with CMR.
      One novel finding is correlation analysis, in which we observed excellent correlations between fully automated 3DE and CMR. However, we found that there was moderate to severe between-study heterogeneity. The lowest correlation was for LVESV, perhaps relating to incorrect number and small study size, and omitting this study did not strongly affect overall results.
      To the best of our knowledge, this is the first study to investigate accuracy of LV volumes and LVEF measurements that incorporated two types of fully automated quantification software. Although both still underestimated LV volumes against CMR, the bias did not exceed 20 mL (Graphical Abstract). The LVEF derived by 3DE was similar to the corresponding values measured by CMR. We observed high feasibility of the software analysis as well as excellent correlations of LV parameters between the two techniques. There were no differences in the software whether it was the previous generation HM or latest generation DHM compared to CMR. In addition, there was no difference in the type of the analysis whether it was fully automated or manually edited in LVEDV and LVESV, but little difference in LVEF (−1.4 % and 1.5 % respectively) compared to CMR.
      To summarize the clinical implication of our study, 3DE derived LV volumes and LVEF were comparable to CMR counterparts. There were no significant associations between publication year, type of software, or type of analysis and bias between the two techniques. We believe routine adoption for clinical settings in adult population should be practical.

      Limitations

      There are several limitations that should be addressed in this study. First, the number of comparative articles using fully automated left chamber quantification software was still small, and more studies should be included to obtain consistent findings. There were also few studies (n = 4) that incorporated LA volume data; thus, we did not analyze LA parameters in this study. Second, like other meta-analyses, results could be influenced by publication bias. Although a funnel plot and Egger's test showed no significant asymmetry, the sensitivity of these tests is relatively low. Third, each study applied different LV border thresholds, and some studies did not provide border setting information. Therefore, observed results were not obtained with fixed threshold settings for measurements. Further studies should be required to determine the best LV border setting. Fourth, only 2 studies included patients with congenital heart disease while others either did not include or did not specify if they had included these patients. Patients with dilated heart were included in the study. Therefore although many studies did not specify how many patients of severe valvular heart disease or extremely poor function were included, they could be included in the study if they did not have poor endocardial visualization on 2D echocardiography of ≥3 contiguous segments using a 17-segment model. Fifth, our meta-analysis is based on the results under the assumption of normal distribution, but because we could not obtain the original data of 12 articles, we therefore failed to test the assumption of a normal distribution. Finally, the software is vendor-dependent; thus, results may not be generalized to datasets of other ultrasound vendors.

      Conclusions

      Although fully automated 3DE left chamber quantification software still underestimated LV volumes, EF showed similar values to CMR. Excellent correlations between the two techniques make fully automated analysis possible, facilitating routine adoption for clinical settings in adult population.
      Online Fig. 1
      Online Fig. 1Outlier and influence analyses were performed to detect outliers/influential cases for mean differences of LVEDV (A), LVESV (B), and LVEF (C) between 3DE and CMR.
      Online Fig. 2
      Online Fig. 2Leave-one-out analyses were performed to detect outliers/influential cases for mean differences of LVEDV (A), LVESV (B), and LVEF (C) between 3DE and CMR.
      Online Fig. 3
      Online Fig. 3Outlier and influence analyses were performed to detect outliers/influential cases for correlations of LVEDV (A), LVESV (B), and LVEF (C) between 3DE and CMR.
      Online Fig. 4
      Online Fig. 4Leave-one-out analyses were performed to detect outliers/influential cases for correlations of LVEDV (A), LVESV (B), and LVEF (C) between 3DE and CMR.

      Declaration of competing interest

      All authors have nothing to disclose.

      Acknowledgments

      None.

      Author contributions

      VCW and MT contributed to conception and design of the study. VCW and TK organized the database. VCW, MT performed the statistical analysis. VCW and MT wrote the first draft of the manuscript. VCW, PHC, MT wrote sections of the manuscript. All authors contributed to manuscript revision, read, and approved the submitted version.

      References

        • Lang R.M.
        • Addetia K.
        • Narang A.
        • Mor-Avi V.
        3-dimensional echocardiography latest developments and future directions.
        JACC Cardiovasc Imaging. 2018; 11: 1854-1878
        • Medvedofsky D.
        • Mor-Avi V.
        • Amzulescu M.
        • Fernández-Golfín C.
        • Hinojar R.
        • Monaghan M.J.
        • et al.
        Three-dimensional echocardiographic quantification of the left-heart chambers using an automated adaptive analytics algorithm: multicentre validation study.
        Eur Heart J Cardiovasc Imaging. 2018; 19: 47-58
        • Barbieri A.
        • Bursi F.
        • Camaioni G.
        • Maisano A.
        • Imberti J.F.
        • Albini A.
        • et al.
        Echocardiographic left ventricular mass assessment: correlation between 2D-derived linear dimensions and 3-dimensional automated, machine learning-based methods in unselected patients.
        J Clin Med. 2021; 10: 1279
        • Kitano T.
        • Nabeshima Y.
        • Otsuji Y.
        • Takeuchi M.
        Accuracy of left ventricular volumes and ejection fraction measurements by contemporary three-dimensional echocardiography with semi- and fully automated software: systematic review and meta-analysis of 1,881 subjects.
        J Am Soc Echocardiogr. 2019; 32: 1105-1115
        • Généreux P.
        • Stone G.W.
        • O'Gara P.T.
        • Marquis-Gravel G.
        • Redfors B.
        • Giustino G.
        • et al.
        Natural history, diagnostic approaches, and therapeutic strategies for patients with asymptomatic severe aortic stenosis.
        J Am Coll Cardiol. 2016; 19: 2263-2288
        • Egger M.
        • Davey Smith G.
        • Schneider M.
        • Minder C.
        Bias in meta-analysis detected by a simple, graphical test.
        Br Med J. 1997; 315: 629-634
        • Downs S.H.
        • Black N.
        The feasibility of creating a checklist for the assessment of the methodological quality both of randomised and nonrandomised studies of health care interventions.
        J Epidemiol Comm Health. 1998; 52: 377-384
        • Tsang W.
        • Salgo I.S.
        • Medvedofsky D.
        • Takeuchi M.
        • Prater D.
        • Weinert L.
        • et al.
        Transthoracic 3D echocardiographic left heart chamber quantification using an automated adaptive analytics algorithm.
        JACC Cardiovasc Imaging. 2016; 9: 769-782
        • Yang L.T.
        • Nagata Y.
        • Otani K.
        • Kado Y.
        • Otsuji Y.
        • Takeuchi M.
        Feasibility of one-beat real-time full-volume three-dimensional echocardiography for assessing left ventricular volumes and deformation parameters.
        J Am Soc Echocardiogr. 2016; 29: 853-860
        • Levy F.
        • Schouver E.D.
        • Iacuzio L.
        • Civaia F.
        • Rusek S.
        • Dommerc C.
        • et al.
        Performance of new automated transthoracic three-dimensional echocardiographic software for left ventricular volumes and function assessment in routine clinical practice: comparison with 3 tesla cardiac magnetic resonance.
        Arch Cardiovasc Dis. 2017; 110: 580-589
        • Tamborini G.
        • Piazzese C.
        • Lang R.M.
        • Muratori M.
        • Chiorino E.
        • Mapelli M.
        • et al.
        Feasibility and accuracy of automated software for transthoracic three-dimensional left ventricular volume and function analysis: comparisons with two-dimensional echocardiography, three-dimensional transthoracic manual method, and cardiac magnetic resonance imaging.
        J Am Soc Echocardiogr. 2017; 30: 1049-1058
        • Żygadło A.
        • Kaźnica-Wiatr M.
        • Błaut-Jurkowska J.
        • Knap K.
        • Lenart-Migdalska A.
        • Smaś-Suska M.
        • et al.
        Evaluation of the clinical suitability of automated left ventricle's fraction and volume measurements in 3-dimensional echocardiography compared to values obtained in magnetic resonance imaging (pilot study).
        Pol Med J. 2017; 43: 154-157
        • Barletta V.
        • Hinojar R.
        • Carbonell A.
        • González-Gómez A.
        • Fabiani I.
        • Di Bello V.
        • et al.
        Three-dimensional full automated software in the evaluation of the left ventricle function: from theory to clinical practice.
        Int J Cardiovasc Imaging. 2018; 34: 1205-1213
        • Levy F.
        • Marechaux S.
        • Iacuzio L.
        • Schouver E.D.
        • Castel A.L.
        • Toledano M.
        • et al.
        Quantitative assessment of primary mitral regurgitation using left ventricular volumes obtained with new automated three-dimensional transthoracic echocardiographic software: a comparison with 3-tesla cardiac magnetic resonance.
        Arch Cardiovasc Dis. 2018; 111: 507-517
        • Narang A.
        • Mor-Avi V.
        • Prado A.
        • Volpato V.
        • Prater D.
        • Tamborini G.
        • et al.
        Machine learning based automated dynamic quantification of left heart chamber volumes.
        Eur Heart J Cardiovasc Imaging. 2019; 20: 541-549
        • Volpato V.
        • Mor-Avi V.
        • Narang A.
        • Prater D.
        • Gonçalves A.
        • Tamborini G.
        • et al.
        Automated, machine learning-based, 3D echocardiographic quantification of left ventricular mass.
        Echocardiography. 2019; 36: 312-319
        • Wu V.C.
        • Kitano T.
        • Nabeshima Y.
        • Otani Y.
        • Chu P.H.
        • Takeuchi M.
        Optimal threshold of three-dimensional echocardiographic fully automated software for quantification of left ventricular volumes and ejection fraction: comparison with cardiac magnetic resonance disk-area summation method and feature tracking method.
        PLoS One. 2019; 14e0211154
        • Levy F.
        • Iacuzio L.
        • Schouver E.D.
        • Essayagh B.
        • Civaia F.
        • Dommerc C.
        • et al.
        Performance of a new fully automated transthoracic three-dimensional echocardiographic software for quantification of left cardiac chamber size and function: comparison with 3 tesla cardiac magnetic resonance.
        J Clin Ultrasound. 2019; 47: 546-554
        • Italiano G.
        • Tamborini G.
        • Fusini L.
        • Mantegazza V.
        • Doldi M.
        • Celeste F.
        • et al.
        Feasibility and accuracy of the automated software for dynamic quantification of left ventricular and atrial volumes and function in a large unselected population.
        J Clin Med. 2021; 10: 5030
        • Shimada Y.J.
        • Shiota T.
        A meta-analysis and investigation for the source of bias of left ventricular volumes and function by three-dimensional echocardiography in comparison with magnetic resonance imaging.
        Am J Cardiol. 2011; 107: 126-138
        • Dorosz J.L.
        • Lezotte D.C.
        • Weitzenkamp D.A.
        • Allen L.A.
        • Salcedo E.E.
        Performance of 3-dimensional echocardiography in measuring left ventricular volumes and ejection fraction: a systematic review and meta-analysis.
        J Am Coll Cardiol. 2012; 59: 1799-1808
        • Rigolli M.
        • Anandabaskaran S.
        • Christiansen J.P.
        • Whalley G.A.
        Bias associated with left ventricular quantification by multimodality imaging - a systematic review and meta-analysis.
        Open Heart. 2016; 3e000388