TY - JOUR
AU - Takeuchi,, Masaaki
AB - Abstract Aims The aim of this study was to determine the accuracy and reproducibility of a novel, fully automated 3D echocardiography (3DE) right ventricular (RV) quantification software compared with cardiac magnetic resonance (CMR) and semi-automated 3DE RV quantification software. Methods and results RV volumes and the RV ejection fraction (RVEF) were measured using a fully automated software (Philips), a semi-automated software (TomTec), and CMR in 100 patients who had undergone both CMR and 3DE examinations on the same day. The feasibility of the fully automated software was 91%. Although the fully automated software, without any manual editing, significantly underestimated RV end-diastolic volume (bias: −12.6 mL, P < 0.001) and stroke volume (−5.1 mL, P < 0.001) compared with CMR, there were good correlations between the two modalities (r = 0.82 and 0.78). No significant differences in RVEF between the fully automated software and CMR were observed, and there was a fair correlation (r = 0.72). The RVEF determined by the semi-automated software was significantly larger than that by CMR or the fully automated software (P < 0.001). The fully automated software had a shorter analysis time compared with the semi-automated software (15 s vs. 120 s, P < 0.001) and had a good reproducibility. Conclusion A novel, fully automated 3DE RV quantification software underestimated RV volumes but successfully approximated RVEF when compared with CMR. No inferiority of this software was observed when compared with the semi-automated software. Rapid analysis and higher reproducibility also support the routine adoption of this method in the daily clinical workflow. right ventricle, right ventricular volume and ejection fraction, 3D echocardiography, cardiac magnetic resonance, validation Introduction The right ventricle plays a pivotal role in cardiac function.1,2 Evaluation of right ventricular (RV) function is increasingly popular to diagnose RV dysfunction and predict future adverse outcomes.3–6 Since the right ventricle has an irregular shape, 3D analysis is theoretically more accurate and reliable than 2D analysis.7 Although cardiac magnetic resonance (CMR) is an established method to measure RV volumes and RV ejection fraction (RVEF),8,9 3D echocardiography (3DE) is another method of choice due to versatility, availability, and lower cost. Previous 3DE studies have reported the reference values of RV volumes and RVEF, evaluation of RV dysfunction, and their prognostic values.10–14 Although 3DE RV quantification software has steadily advanced,15–20 the software still requires manual input, which may cause measurement variabilities that will not be negligible. Reliable manual editing heavily depends on the examiner’s knowledge and expertise. The adoption of a fully automated RV quantification software is the only way to resolve this problem. Recently, a fully automated 3DE RV quantification software using machine learning algorithms has been developed, and its accuracy was validated against CMR.21 However, the measurement accuracy of a solely fully automated approach has not been determined, and direct comparison between the currently available semi-automated 3DE RV quantification software and the novel software has not been conducted. Accordingly, the aims of this study were to (i) determine the accuracy of a fully automated 3DE RV quantification software to measure RV volumes and RVEF against CMR as a reference, (ii) compare measurement results with a semi-automated 3DE RV quantification software, (iii) evaluate reproducibility, and (iv) measure the time required for the analysis. Methods Study population This is an observational cross-sectional study from a single centre. We retrospectively selected patients who had undergone clinically indicated CMR and who were also willing to undergo 3DE examination on the same day from 1 January 2018 to 23 July 2019 in our hospital. Patient selection was solely related to the clinical indication of CMR, and no patients were excluded due to poor echocardiography image quality. Among 131 patients who met the inclusion criteria, we selected 100 patients whose 3DE examination had been performed with a specific manufacturer’s ultrasound machine and equipment (Epic 7G, Philips Healthcare, Andover, MA, USA). The study protocol was approved by the ethics committee of the university, and informed consent was waived due to the retrospective nature of the analysis. Echocardiography After acquiring standard 2D and Doppler echocardiography datasets using a X5-1 matrix transducer (Philips Healthcare), a full-volume 3DE dataset that encompassed the entirety of the left ventricle and the left atrium was acquired by the apical approach using a specific one-beat acquisition mode (HMQ). The acquisition was performed several times, and each dataset was stored on the ultrasound machine. Subsequently, the transducer position was moved towards the lateral side of the chest wall, and a 2D RV focused view was obtained. After ensuring this was the best view, 3DE HMQ mode was reactivated to acquire a one-beat full-volume 3DE dataset focusing on the right ventricle and the right atrium, which was performed several times. Lateral and elevation width of the image sector size was reduced to be as narrow as possible keeping with two orthogonal 2D echocardiography images that encompassed the whole part of the right ventricle. The datasets were stored on the ultrasound machine. Finally, the image datasets were transferred to a separate workstation for later analysis. 3DE analysis: 3D auto RV Image quality of 3DE datasets was assessed subjectively according to the completeness of RV endocardial visualization on three 2D images extracted from 3DE datasets (apical four-chamber view, coronal view, and basal short-axis view) and categorized as good, fair, or poor. RV focused one-beat 3DE full-volume datasets were analysed using a novel, fully automated RV quantification software (3D Auto RV, Philips Medical Systems) that detects RV endocardial surfaces using artificial intelligence, which consists of knowledge-based identification of initial global shape and RV chamber orientation, followed by 3D speckle tracking analysis throughout one cardiac cycle.21 After starting the programme, the software automatically performed a view adjustment and constructed a 3D endocardial cast of the right ventricle at the end-diastole by the Heart Model segmentation algorithm. Subsequently, the software performed 3D speckle tracking analysis on the RV endocardial border and provided RV volumes curves, from which the RV end-diastolic volume (RVEDV), RV end-systolic volume (RVESV), RV stroke volume (RVSV), and RVEF were determined (Figure 1). Figure 1 Open in new tabDownload slide Process of fully automated right ventricular volumetric analysis. After retrieving 3DE datasets aimed for the right ventricle (A) and starting the software, the software automatically determined the right ventricular border at the end-diastole (B) and end-systole (C), and provided a 3D RV cast and volumetric data within 15 s. Figure 1 Open in new tabDownload slide Process of fully automated right ventricular volumetric analysis. After retrieving 3DE datasets aimed for the right ventricle (A) and starting the software, the software automatically determined the right ventricular border at the end-diastole (B) and end-systole (C), and provided a 3D RV cast and volumetric data within 15 s. When required, end-diastolic RV endocardial contours were adjusted manually on both 2D long- and short-axis views, and new speckle tracking analysis on the corrected RV contour was conducted after clicking the ‘RV Re-tracking’ label. If some part of the RV wall still did not track adequately, a manual correction of RV endocardial border in the end-systolic frame was performed. These corrections were continuously updated on the 3D RV cast and then propagated to all of the other frames of the cardiac cycle using the derived tracking information. The software sometimes made an erroneous 3D RV cast (Supplementary data online, Figure S1). When this issue occurred, we clicked the ‘View Adjustment’ label and adjusted four specific anatomical landmarks to the appropriate positions on the corresponding long-axis views. After clicking the ‘Tracking Revision’ label, the software made a new 3D cast, taking into consideration the updated anatomical information. If the software still made an erroneous 3D cast, we discarded the analysis. To determine the accuracy of the fully automated software, we obtained RV volumes and RVEF results using the fully automated method (no manual correction after automatic determination of RV contour). For comparison, we also obtained the results after manual editing at the end-diastolic frame and sometimes at the end-systolic frame in every patient (a manual editing method). 3DE analysis: 4D RV function 2 The same 3DE datasets were exported in DICOM format to another workstation to measure RV volumes and RVEF using a commercially available semi-automated vendor-independent 3DE RV quantification software (RV-Function 2, TomTec Imaging Systems GmbH, Unterschleissheim, Germany). The methodologies for the RV quantification have been previously reported.13,17 After registration of some anatomical landmarks, the software generated a 3D RV cast, with which we performed manual editing of the RV endocardial border for both the end-diastolic and end-systolic images, thus resulting in the calculation of the RV volumes and RVEF. CMR acquisition CMR imaging was performed with a 3T scanner (Discovery 750 W, GE Healthcare Milwaukee, WI, USA) with a phased-array cardiovascular coil. Retrospective electrocardiographically gated localizing spin-echo sequences were used to identify the long axis of the heart. Steady-state free precession (SSFP) dynamic gradient-echo cine loops were acquired using retrospective electrocardiographic gating and parallel imaging techniques during 10-s to 15-s breath-holds with the following general parameters: 8 mm slice thickness of the imaging planes, 40 × 40 cm field of view, 200 × 160 scan matrix, 50° flip angle, 3.8/1.7 ms repetition/echo times, and 20–30 reconstructed cardiac phases. CMR analysis CMR RV volumes were measured from multiple short-axis SSFP images using an analytical software (Segment version 2.2, Medviso, Lund, Sweden), and the disk-area summation method was used for the calculation of the RVEDV and RVESV.9 RVSV was determined by RVEDV − RVESV. The RVEF was measured by standard formula. Reproducibility Intra-observer variability of the RV volumes and RVEF were assessed by having the examiner repeat the measurements using three analytical methods at a 2-week interval for 20 randomly selected patients, and inter-observer variabilities were determined by having a second examiner perform these measurements in the same 20 patients. Test–retest variabilities were assessed by another 3DE dataset that was acquired at a different time point during the same examinations. The intra- and inter-observer and test–retest variability values were calculated as the absolute differences between the corresponding two measurements in percentages of their mean and intraclass correlation. Time required for analysis Analysis time was also measured in the same 20 patients who were used for reproducibility analysis. After selecting image dataset, we calculated the time from starting the software to obtaining the final results. Statistical analysis Continuous data are presented as the mean ± standard deviation or median and interquartile range (IQR, 25th percentile–75th percentile), according to the data distribution. Categorical values are expressed as numbers or percentages. Friedman’s analysis with post hoc comparison was performed to compare the values among four groups. A linear correlation and Bland–Altman analysis were performed to determine the r value, bias, and 95% confidence interval. A P < 0.05 was considered significant. All statistical analyses were performed using commercial software (SPSS version 24, Chicago, IL, USA; R version 3.4.3, The R Foundation for Statistical Computing, Vienna, Austria; Prism 8.1.2, GraphPad Software, Inc. San Diego, CA, USA). Results Study subjects Table 1 represents the clinical characteristics and CMR results in the study subjects. RV image quality analysis was good in 16 patients (16%), fair in 46 patients (46%), poor in 38 patients (38%), respectively. The median volume rate of 3DE datasets was 23/s. The fully automated RV software made erroneous RV casts in 20 cases, thus, requiring a view adjustment. Among them, image quality was poor in 17 patients. 3DE datasets were acquired from unusual locations (more medial site near the sternum or subcostal approach) in 11 out of 20 patients. Even after the view adjustment, the software still provided an erroneous RV cast in seven patients. In addition, there were no 3DE datasets because of extremely poor images in one patient and the lower volume rate precluded the analysis in one patient, resulting in feasibility being 91%. The corresponding feasibility with the semi-automated software (TomTec) was 97% (97/100). RV volumetric analysis using CMR was not possible in six patients due to respiratory artefacts or an electrocardiography problem (feasibility: 94%). Thus, direct comparison among the four methods was possible in 87 patients. Table 1 Clinical characteristics and CMR results in the study patients (n = 100) Variables Age (years) 67 ± 14 Male/female 62/38 Body surface area (/m2) 1.58 ± 0.19 Clinical diagnosis  Ischaemic heart disease 36 (36%)  Secondary cardiomyopathy 26 (26%)  Valvular heart disease 9 (9%)  Pulmonary hypertension 7 (7%)  Dilated cardiomyopathy 5 (5%)  Cardiac tumour 4 (4%)  Hypertrophic cardiomyopathy 2 (2%)  Others 8 (8%) Heart rate (bpm) 69 ± 14 Systolic blood pressure (mmHg) 130 ± 24 Diastolic blood pressure (mmHg) 72 ± 13 CMR results  Left ventricular end-diastolic volume (mL) 134 (105–203)  Left ventricular end-systolic volume (mL) 80 (49–144)  Left ventricular ejection fraction (%) 40.8 (26.2–53.4)  Right ventricular end-diastolic volume (mL) 104 (88–131)  Right ventricular end-diastolic volume (mL) 57 (43–82)  Right ventricular ejection fraction (%) 43.3 (36.1–51.5) Variables Age (years) 67 ± 14 Male/female 62/38 Body surface area (/m2) 1.58 ± 0.19 Clinical diagnosis  Ischaemic heart disease 36 (36%)  Secondary cardiomyopathy 26 (26%)  Valvular heart disease 9 (9%)  Pulmonary hypertension 7 (7%)  Dilated cardiomyopathy 5 (5%)  Cardiac tumour 4 (4%)  Hypertrophic cardiomyopathy 2 (2%)  Others 8 (8%) Heart rate (bpm) 69 ± 14 Systolic blood pressure (mmHg) 130 ± 24 Diastolic blood pressure (mmHg) 72 ± 13 CMR results  Left ventricular end-diastolic volume (mL) 134 (105–203)  Left ventricular end-systolic volume (mL) 80 (49–144)  Left ventricular ejection fraction (%) 40.8 (26.2–53.4)  Right ventricular end-diastolic volume (mL) 104 (88–131)  Right ventricular end-diastolic volume (mL) 57 (43–82)  Right ventricular ejection fraction (%) 43.3 (36.1–51.5) Continuous data are expressed as mean ± standard deviation or median and interquartile range. Open in new tab Table 1 Clinical characteristics and CMR results in the study patients (n = 100) Variables Age (years) 67 ± 14 Male/female 62/38 Body surface area (/m2) 1.58 ± 0.19 Clinical diagnosis  Ischaemic heart disease 36 (36%)  Secondary cardiomyopathy 26 (26%)  Valvular heart disease 9 (9%)  Pulmonary hypertension 7 (7%)  Dilated cardiomyopathy 5 (5%)  Cardiac tumour 4 (4%)  Hypertrophic cardiomyopathy 2 (2%)  Others 8 (8%) Heart rate (bpm) 69 ± 14 Systolic blood pressure (mmHg) 130 ± 24 Diastolic blood pressure (mmHg) 72 ± 13 CMR results  Left ventricular end-diastolic volume (mL) 134 (105–203)  Left ventricular end-systolic volume (mL) 80 (49–144)  Left ventricular ejection fraction (%) 40.8 (26.2–53.4)  Right ventricular end-diastolic volume (mL) 104 (88–131)  Right ventricular end-diastolic volume (mL) 57 (43–82)  Right ventricular ejection fraction (%) 43.3 (36.1–51.5) Variables Age (years) 67 ± 14 Male/female 62/38 Body surface area (/m2) 1.58 ± 0.19 Clinical diagnosis  Ischaemic heart disease 36 (36%)  Secondary cardiomyopathy 26 (26%)  Valvular heart disease 9 (9%)  Pulmonary hypertension 7 (7%)  Dilated cardiomyopathy 5 (5%)  Cardiac tumour 4 (4%)  Hypertrophic cardiomyopathy 2 (2%)  Others 8 (8%) Heart rate (bpm) 69 ± 14 Systolic blood pressure (mmHg) 130 ± 24 Diastolic blood pressure (mmHg) 72 ± 13 CMR results  Left ventricular end-diastolic volume (mL) 134 (105–203)  Left ventricular end-systolic volume (mL) 80 (49–144)  Left ventricular ejection fraction (%) 40.8 (26.2–53.4)  Right ventricular end-diastolic volume (mL) 104 (88–131)  Right ventricular end-diastolic volume (mL) 57 (43–82)  Right ventricular ejection fraction (%) 43.3 (36.1–51.5) Continuous data are expressed as mean ± standard deviation or median and interquartile range. Open in new tab Comparison of RV volumes and RVEF among four methods Figure 2 depicts RV volumes, RVEF, and RVSV using four different methods. The median value of CMR determined that RVEDV was 105 mL. The fully automated method and the semi-automated software significantly underestimated the RVEDV compared with CMR. However, there were no significant differences in RVEDV between CMR and the manual editing method (P = 0.77). The same trend was also observed in RVESV. The median value of RVEF using CMR was 43.4%. The semi-automated software had a significantly higher RVEF than CMR (P = 0.003). There were no significant differences in RVEF between CMR and the other two echocardiography methods. The median value of RVSV using CMR was 45 mL. The fully automated method and the semi-automated software significantly underestimated RVSV compared with CMR. However, there were no significant differences in RVSV between CMR and the manual editing method (P = 1.000). No significant differences in RVEDV (P = 0.12) and RVSV (P = 1.000) were noted between the fully automated method and the semi-automated software. However, RVESV determined by using the semi-automated software was significantly lower than that obtained by using the fully automated method (P = 0.002), resulting in a significant higher RVEF by semi-automated software compared with fully automated method (P < 0.001). Figure 2 Open in new tabDownload slide A comparison of right ventricular parameters for the four methods. (A) Right ventricular end-diastolic volume (RVEDV); (B) right ventricular end-systolic volume (RVESV); (C) right ventricular ejection fraction (RVEF); and (D) right ventricular stroke volume (RVSV). 3D Auto RV edit, 3D Auto RV analysis with manual editing; CMR, cardiac magnetic resonance. Figure 2 Open in new tabDownload slide A comparison of right ventricular parameters for the four methods. (A) Right ventricular end-diastolic volume (RVEDV); (B) right ventricular end-systolic volume (RVESV); (C) right ventricular ejection fraction (RVEF); and (D) right ventricular stroke volume (RVSV). 3D Auto RV edit, 3D Auto RV analysis with manual editing; CMR, cardiac magnetic resonance. Correlation and Bland–Altman analysis between 3DE and CMR Figure 3 presents the linear correlation and Bland–Altman analysis for RVEDV using three 3DE RV quantification methods against CMR as a reference. There were good correlations of RVEDV between CMR and the three 3DE RV quantification methods. However, compared with the semi-automated software, bias decreased and the slope of the equation become steeper for the fully automated method and the manual editing method. The same trend was also observed in RVESV. Regarding RVEF, there was a good correlation between CMR and the semi-automated software (r = 0.80). While the corresponding value using the fully automated method was 0.72, the value further improved to 0.87 when the manual editing method was used (Figure 4). There was a modest correlation of RVSV between CMR and the semi-automated software (r = 0.67). With use of the fully automated method, the bias became less and the slope of the equation became steeper compared with the semi-automated software (Figure 5). Figure 3 Open in new tabDownload slide A linear correlation and Bland–Altman analysis of right ventricular end-diastolic volume between the echocardiographic method and cardiac magnetic resonance. (A) Semi-automated method vs. CMR. (B) Fully automated method vs. CMR. (C) Manual editing method vs. CMR. 3D Auto RV edit, 3D Auto RV analysis with manual editing; CMR, cardiac magnetic resonance; TomTec, RV Function 2. Figure 3 Open in new tabDownload slide A linear correlation and Bland–Altman analysis of right ventricular end-diastolic volume between the echocardiographic method and cardiac magnetic resonance. (A) Semi-automated method vs. CMR. (B) Fully automated method vs. CMR. (C) Manual editing method vs. CMR. 3D Auto RV edit, 3D Auto RV analysis with manual editing; CMR, cardiac magnetic resonance; TomTec, RV Function 2. Figure 4 Open in new tabDownload slide A linear correlation and Bland–Altman analysis of right ventricular ejection fraction between the echocardiographic method and cardiac magnetic resonance. (A) Semi-automated method vs. CMR. (B) Fully automated method vs. CMR. (C) Manual editing method vs. CMR. 3D Auto RV edit, 3D Auto RV analysis with manual editing; CMR, cardiac magnetic resonance; TomTec, RV Function 2. Figure 4 Open in new tabDownload slide A linear correlation and Bland–Altman analysis of right ventricular ejection fraction between the echocardiographic method and cardiac magnetic resonance. (A) Semi-automated method vs. CMR. (B) Fully automated method vs. CMR. (C) Manual editing method vs. CMR. 3D Auto RV edit, 3D Auto RV analysis with manual editing; CMR, cardiac magnetic resonance; TomTec, RV Function 2. Figure 5 Open in new tabDownload slide A linear correlation and Bland–Altman analysis of right ventricular stroke volume between three echocardiographic methods and cardiac magnetic resonance. (A) Semi-automated method vs. CMR. (B) Fully automated method vs. CMR. (C) Manual editing method vs. CMR. 3D Auto RV edit, 3D Auto RV analysis with manual editing; CMR, cardiac magnetic resonance; TomTec, RV Function 2. Figure 5 Open in new tabDownload slide A linear correlation and Bland–Altman analysis of right ventricular stroke volume between three echocardiographic methods and cardiac magnetic resonance. (A) Semi-automated method vs. CMR. (B) Fully automated method vs. CMR. (C) Manual editing method vs. CMR. 3D Auto RV edit, 3D Auto RV analysis with manual editing; CMR, cardiac magnetic resonance; TomTec, RV Function 2. Effect of image quality on bias and r values Table 2 shows bias and the r value of RV volumes, RVEF, and RVSV of each of the 3DE RV quantification methods against CMR according to the image quality. Image quality heavily affected the bias and r values of all four RV parameters using semi-automated software (e.g. good image quality − less bias and higher r value). Overall, a similar trend was also observed by the fully automated software, but the trend was not consistent in every RV parameter. Table 2 Effect of image quality on bias and r value of RV volumes and RVEF between echocardiography quantification method and CMR Semi-automated (TomTec) Fully automated method Manual editing method Bias r Bias r Bias r RVEDV  Overall (n = 87) −20.0 0.73 −12.6 0.82 −2.9 0.83  Good (n = 16) −12.5 0.76 −11.4 0.83 2.8 0.86  Fair (n = 42) −18.7 0.77 −11.6 0.78 −1.8 0.80  Poor (n = 29) −25.9 0.65 −14.8 0.85 −7.8 0.86 RVESV  Overall (n = 87) −13.6 0.79 −7.5 0.82 −2.4 0.87  Good (n = 16) −10.5 0.87 −11.3 0.91 0.5 0.92  Fair (n = 42) −12.5 0.78 −5.7 0.79 −1.3 0.84  Poor (n = 29) −16.9 0.70 −8.1 0.77 −5.8 0.84 RVEF  Overall (n = 87) 2.4 0.80 −0.3 0.72 0.8 0.87  Good (n = 16) 1.9 0.93 2.3 0.84 0.4 0.87  Fair (n = 42) 2.1 0.67 −1.3 0.58 0.2 0.80  Poor (n = 29) 3.1 0.81 −0.3 0.76 1.7 0.89 RVSV  Overall (n = 87) −6.3 0.67 −5.1 0.78 −0.5 0.79  Good (n = 16) −2.0 0.74 −0.1 0.78 2.3 0.75  Fair (n = 42) −6.2 0.70 −5.9 0.67 −0.5 0.71  Poor (n = 29) −9.0 0.70 −6.7 0.91 −2.1 0.90 Semi-automated (TomTec) Fully automated method Manual editing method Bias r Bias r Bias r RVEDV  Overall (n = 87) −20.0 0.73 −12.6 0.82 −2.9 0.83  Good (n = 16) −12.5 0.76 −11.4 0.83 2.8 0.86  Fair (n = 42) −18.7 0.77 −11.6 0.78 −1.8 0.80  Poor (n = 29) −25.9 0.65 −14.8 0.85 −7.8 0.86 RVESV  Overall (n = 87) −13.6 0.79 −7.5 0.82 −2.4 0.87  Good (n = 16) −10.5 0.87 −11.3 0.91 0.5 0.92  Fair (n = 42) −12.5 0.78 −5.7 0.79 −1.3 0.84  Poor (n = 29) −16.9 0.70 −8.1 0.77 −5.8 0.84 RVEF  Overall (n = 87) 2.4 0.80 −0.3 0.72 0.8 0.87  Good (n = 16) 1.9 0.93 2.3 0.84 0.4 0.87  Fair (n = 42) 2.1 0.67 −1.3 0.58 0.2 0.80  Poor (n = 29) 3.1 0.81 −0.3 0.76 1.7 0.89 RVSV  Overall (n = 87) −6.3 0.67 −5.1 0.78 −0.5 0.79  Good (n = 16) −2.0 0.74 −0.1 0.78 2.3 0.75  Fair (n = 42) −6.2 0.70 −5.9 0.67 −0.5 0.71  Poor (n = 29) −9.0 0.70 −6.7 0.91 −2.1 0.90 RVED(S)V, right ventricular end-diastolic (end-systolic) volume; RVEF, right ventricular ejection fraction; RVSV, right ventricular stroke volume. Open in new tab Table 2 Effect of image quality on bias and r value of RV volumes and RVEF between echocardiography quantification method and CMR Semi-automated (TomTec) Fully automated method Manual editing method Bias r Bias r Bias r RVEDV  Overall (n = 87) −20.0 0.73 −12.6 0.82 −2.9 0.83  Good (n = 16) −12.5 0.76 −11.4 0.83 2.8 0.86  Fair (n = 42) −18.7 0.77 −11.6 0.78 −1.8 0.80  Poor (n = 29) −25.9 0.65 −14.8 0.85 −7.8 0.86 RVESV  Overall (n = 87) −13.6 0.79 −7.5 0.82 −2.4 0.87  Good (n = 16) −10.5 0.87 −11.3 0.91 0.5 0.92  Fair (n = 42) −12.5 0.78 −5.7 0.79 −1.3 0.84  Poor (n = 29) −16.9 0.70 −8.1 0.77 −5.8 0.84 RVEF  Overall (n = 87) 2.4 0.80 −0.3 0.72 0.8 0.87  Good (n = 16) 1.9 0.93 2.3 0.84 0.4 0.87  Fair (n = 42) 2.1 0.67 −1.3 0.58 0.2 0.80  Poor (n = 29) 3.1 0.81 −0.3 0.76 1.7 0.89 RVSV  Overall (n = 87) −6.3 0.67 −5.1 0.78 −0.5 0.79  Good (n = 16) −2.0 0.74 −0.1 0.78 2.3 0.75  Fair (n = 42) −6.2 0.70 −5.9 0.67 −0.5 0.71  Poor (n = 29) −9.0 0.70 −6.7 0.91 −2.1 0.90 Semi-automated (TomTec) Fully automated method Manual editing method Bias r Bias r Bias r RVEDV  Overall (n = 87) −20.0 0.73 −12.6 0.82 −2.9 0.83  Good (n = 16) −12.5 0.76 −11.4 0.83 2.8 0.86  Fair (n = 42) −18.7 0.77 −11.6 0.78 −1.8 0.80  Poor (n = 29) −25.9 0.65 −14.8 0.85 −7.8 0.86 RVESV  Overall (n = 87) −13.6 0.79 −7.5 0.82 −2.4 0.87  Good (n = 16) −10.5 0.87 −11.3 0.91 0.5 0.92  Fair (n = 42) −12.5 0.78 −5.7 0.79 −1.3 0.84  Poor (n = 29) −16.9 0.70 −8.1 0.77 −5.8 0.84 RVEF  Overall (n = 87) 2.4 0.80 −0.3 0.72 0.8 0.87  Good (n = 16) 1.9 0.93 2.3 0.84 0.4 0.87  Fair (n = 42) 2.1 0.67 −1.3 0.58 0.2 0.80  Poor (n = 29) 3.1 0.81 −0.3 0.76 1.7 0.89 RVSV  Overall (n = 87) −6.3 0.67 −5.1 0.78 −0.5 0.79  Good (n = 16) −2.0 0.74 −0.1 0.78 2.3 0.75  Fair (n = 42) −6.2 0.70 −5.9 0.67 −0.5 0.71  Poor (n = 29) −9.0 0.70 −6.7 0.91 −2.1 0.90 RVED(S)V, right ventricular end-diastolic (end-systolic) volume; RVEF, right ventricular ejection fraction; RVSV, right ventricular stroke volume. Open in new tab Time of analysis The median time of analysis using the fully automated method (14 s; IQR 13–14 s) was significantly shorter than that using the semi-automated software (121 s; IQR 104–155 s, P < 0.001) or the manual editing method (100 s; IQR 86–114 s, P < 0.001). No significant differences in analysis time were noted between the semi-automated software and the manual editing method. Reproducibility Intra- and inter-observer variability for RVEDV, RVESV, and RVEF using the fully automated method was 0%. The corresponding intra- and inter-observer variability of the manual editing method ranged 6.0–7.5% and 5.8–10.2%, respectively, and the values were lower than those using the semi-automated software. Test–retest variability using the fully automated method ranged from 5.5% to 7.8%, and the corresponding values of the manual editing method ranged from 5.7% to 7.6%. These values were less than those obtained by using the semi-automated software (Table 3). Table 3 Reproducibility of three methods for RV volumes and RVEF Method Variables Intra-observer variability Inter-observer variability Test–retest variability % variability, mean ± SD ICC % variability, mean ± SD ICC % variability, mean ± SD ICC Semi-automated software RVEDV 9.8 ± 7.5 0.951 15.3 ± 11.6 0.893 10.5 ± 9.0 0.936 RVESV 13.3 ± 9.4 0.924 14.3 ± 12.0 0.891 14.2 ± 10.6 0.872 RVEF 7.8 ± 5.7 0.743 7.6 ± 6.0 0.655 8.3 ± 8.6 0.633 Fully automated method RVEDV 0 1 0 1 5.5 ± 4.9 0.988 RVESV 0 1 0 1 7.8 ± 7.9 0.985 RVEF 0 1 0 1 6.8 ± 7.4 0.846 Manual editing method RVEDV 7.5 ± 4.5 0.985 8.4 ± 6.1 0.972 7.6 ± 5.0 0.977 RVESV 7.5 ± 7.8 0.980 10.2 ± 6.8 0.949 7.6 ± 7.9 0.982 RVEF 6.0 ± 5.7 0.850 5.8 ± 4.8 0.876 5.7 ± 6.4 0.851 Method Variables Intra-observer variability Inter-observer variability Test–retest variability % variability, mean ± SD ICC % variability, mean ± SD ICC % variability, mean ± SD ICC Semi-automated software RVEDV 9.8 ± 7.5 0.951 15.3 ± 11.6 0.893 10.5 ± 9.0 0.936 RVESV 13.3 ± 9.4 0.924 14.3 ± 12.0 0.891 14.2 ± 10.6 0.872 RVEF 7.8 ± 5.7 0.743 7.6 ± 6.0 0.655 8.3 ± 8.6 0.633 Fully automated method RVEDV 0 1 0 1 5.5 ± 4.9 0.988 RVESV 0 1 0 1 7.8 ± 7.9 0.985 RVEF 0 1 0 1 6.8 ± 7.4 0.846 Manual editing method RVEDV 7.5 ± 4.5 0.985 8.4 ± 6.1 0.972 7.6 ± 5.0 0.977 RVESV 7.5 ± 7.8 0.980 10.2 ± 6.8 0.949 7.6 ± 7.9 0.982 RVEF 6.0 ± 5.7 0.850 5.8 ± 4.8 0.876 5.7 ± 6.4 0.851 Open in new tab Table 3 Reproducibility of three methods for RV volumes and RVEF Method Variables Intra-observer variability Inter-observer variability Test–retest variability % variability, mean ± SD ICC % variability, mean ± SD ICC % variability, mean ± SD ICC Semi-automated software RVEDV 9.8 ± 7.5 0.951 15.3 ± 11.6 0.893 10.5 ± 9.0 0.936 RVESV 13.3 ± 9.4 0.924 14.3 ± 12.0 0.891 14.2 ± 10.6 0.872 RVEF 7.8 ± 5.7 0.743 7.6 ± 6.0 0.655 8.3 ± 8.6 0.633 Fully automated method RVEDV 0 1 0 1 5.5 ± 4.9 0.988 RVESV 0 1 0 1 7.8 ± 7.9 0.985 RVEF 0 1 0 1 6.8 ± 7.4 0.846 Manual editing method RVEDV 7.5 ± 4.5 0.985 8.4 ± 6.1 0.972 7.6 ± 5.0 0.977 RVESV 7.5 ± 7.8 0.980 10.2 ± 6.8 0.949 7.6 ± 7.9 0.982 RVEF 6.0 ± 5.7 0.850 5.8 ± 4.8 0.876 5.7 ± 6.4 0.851 Method Variables Intra-observer variability Inter-observer variability Test–retest variability % variability, mean ± SD ICC % variability, mean ± SD ICC % variability, mean ± SD ICC Semi-automated software RVEDV 9.8 ± 7.5 0.951 15.3 ± 11.6 0.893 10.5 ± 9.0 0.936 RVESV 13.3 ± 9.4 0.924 14.3 ± 12.0 0.891 14.2 ± 10.6 0.872 RVEF 7.8 ± 5.7 0.743 7.6 ± 6.0 0.655 8.3 ± 8.6 0.633 Fully automated method RVEDV 0 1 0 1 5.5 ± 4.9 0.988 RVESV 0 1 0 1 7.8 ± 7.9 0.985 RVEF 0 1 0 1 6.8 ± 7.4 0.846 Manual editing method RVEDV 7.5 ± 4.5 0.985 8.4 ± 6.1 0.972 7.6 ± 5.0 0.977 RVESV 7.5 ± 7.8 0.980 10.2 ± 6.8 0.949 7.6 ± 7.9 0.982 RVEF 6.0 ± 5.7 0.850 5.8 ± 4.8 0.876 5.7 ± 6.4 0.851 Open in new tab Discussion The major findings of this study can be summarized as follows: (i) compared with CMR, a novel 3DE fully automated RV quantification software without any manual editing still significantly underestimated RV volumes with a mean bias of <−13 mL; (ii) there were no significant differences in RVEF between CMR and the fully automated method with a fair correlation (r = 0.72); (iii) there was no inferiority for the measurement accuracy of RV volumes and RVEF using the fully automated method compared with the semi-automated software; (iv) the manual editing method further improved measurement accuracy; and (v) the fully automated method provided rapid analysis time and excellent reproducibility. Previous studies The right ventricle is an irregular shape; thus, accurate analysis requires 3D assessment. 3DE RV quantification software first appeared 10 years ago, and its accuracy has been validated against CMR and multi-detector computed tomography.20 A subsequent study revealed that the software significantly underestimated RV volumes when compared with CMR in adult congenital heart disease patients, which was probably due to the difficulty of endocardial border tracing on the coronal view and some restrictions of manual editing.15 In addition to steady improvements of temporal and spatial resolution of 3DE images, the RV analytical software has also been updated with a more user-friendly interface and improved accuracy. Muraru et al.17 reported that the latest version of semi-automated vendor-independent 3DE RV quantification software had an excellent accuracy (bias for RVEDV, −15 mL; bias for RVEF, 1.4%) and reproducibility compared with CMR. However, the software still required manual input, such as registration of some specific anatomical landmarks and manual corrections of the RV endocardial border on multiple long- and short-axis views at both end-diastolic and end-systolic frames; thus, measurement accuracy and reliability heavily depend on the completeness of manual editing. Manual input also produces observer variabilities. Recently, a fully automated 3DE RV quantification software has been developed using machine learning algorithms. Genovese et al.21 reported that the novel software worked perfectly without any manual editing in 32% of patients but required additional manual editing in the other 68% of patients. Combined results obtained from the fully automated analysis with or without the manual correction had an excellent accuracy (bias for RVEDV, −26 mL; bias for RVEF, 3.3%) when compared with CMR as a reference. The success rate of the fully automated approach was dependent on the image quality. The authors also described that the software had an excellent reproducibility. Current study In this study, we used the same fully automated 3DE RV quantification software and measured RV volumes and RVEF using both the fully automated method and the manual editing method (i.e. manual editing after fully automated analysis) in every patient, and compared these values with CMR and the semi-automated 3DE RV quantification software. The novel, fully automated software made erroneous 3D RV casts in 20 patients. Among them, 17 patients had poor image quality, and 3DE datasets were acquired from an unusual location in 11 patients. Reanalysis after manual correction of anatomical landmarks still provided an erroneous 3D RV cast in seven out of 20 patients. The feasibility of a fully automated approach may be dependent on not only the image quality but also the orientation of the RV delineation (RV apex should be located at the centre of the imaging sector), which is likely because there is less knowledge-based information regarding 3DE RV datasets acquired from unusual locations. Compared with CMR, the fully automated method still significantly underestimated RVEDV. However, the observed bias was <−13 mL, which was within the same range reported by Muraru et al.17 and lower than that reported by Genovese et al.21 There were no significant differences in RVEF between the two methods, and there was a fair correlation (r = 0.72). The fully automated method slightly but significantly underestimated RVSV (5 mL) but showed a good correlation (r = 0.78). Interestingly, image quality heavily affected bias and the correlation coefficients of RV parameters in comparison to CMR when the semi-automated software was used. In contrast, this trend was not consistent using the fully automated method, which was probably because the novel software learned a large amount of RV contour information with the diverse image quality of 3DE datasets. Although the accuracy and reliability of the novel software still depends on the image quality, their contribution was less than those found for the currently available semi-automated software. Compared with the semi-automated 3DE RV quantification software which required extensive manual editing in every patient, the fully automated method had no inferiority for RV volume and RVEF measurements against CMR. The more rapid analysis time and better reproducibility associated with the fully automated method make this software suitable for RV assessment in routine clinical setting. Study limitations There were several study limitations that should be acknowledged. First, this was a single-centre retrospective study, and only a few patients had a very large right ventricle. Thus, further prospective multicentre studies that include a diverse range of RV sizes should be performed to validate our results. Second, the fully automated method was not feasible in 20% of patients. However, the patients were solely selected by a clinical indication of CMR, and 40% of patients had a poor 3DE image quality. 3DE data acquisition from unusual locations, such as a subcostal approach in patients with chronic obstructive lung disease and breast cancer,22 further reduced the success rate of the fully automated analysis. The incorporation of machine learning for 3DE datasets that are acquired from unusual locations might further enhance the feasibility. Third, although bias of the RV volumes decreased compared with previous studies, the fully automated software method still significantly underestimated RV volumes. Fourth, the manual editing method approximated RV volumes, but prolonged analysis times that were similar to the semi-automated software. Last, this is a single vendor product, which will limit its general applicability until other vendors generate similar software. Conclusions A novel, fully automated 3DE RV quantification software still underestimated RV volumes but successfully approximated RVEF when compared with CMR. Rapid analysis and excellent reproducibility supported the routine adoption of this method in the daily clinical workflow. Supplementary data Supplementary data are available at European Heart Journal - Cardiovascular Imaging online. Funding This study was supported by Philips Healthcare. Conflict of interest: M.T. received a research grant from Philips Healthcare and an equipment grant from TomTec Imaging Systems. All other authors have nothing to disclose. References 1 Amsallem M , Mercier O , Kobayashi Y , Moneghetti K , Haddad F. Forgotten no more: a focused update on the right ventricle in cardiovascular disease . JACC Heart Fail 2018 ; 6 : 891 – 903 . Google Scholar Crossref Search ADS PubMed WorldCat 2 Sanz J , Sanchez-Quintana D , Bossone E , Bogaard HJ , Naeije R. Anatomy, function, and dysfunction of the right ventricle: JACC state-of-the-art review . J Am Coll Cardiol 2019 ; 73 : 1463 – 82 . Google Scholar Crossref Search ADS PubMed WorldCat 3 Hamada-Harimura Y , Seo Y , Ishizu T , Nishi I , Machino-Ohtsuka T , Yamamoto M et al. Incremental prognostic value of right ventricular strain in patients with acute decompensated heart failure . Circ Cardiovasc Imaging 2018 ; 11 : e007249 . Google Scholar Crossref Search ADS PubMed WorldCat 4 Houard L , Benaets M-B , de Meester de Ravenstein C , Rousseau MF , Ahn SA , Amzulescu MS et al. Additional prognostic value of 2D right ventricular speckle-tracking strain for prediction of survival in heart failure and reduced ejection fraction: a comparative study with cardiac magnetic resonance . JACC Cardiovasc Imaging 2019 ;doi:10.1016/j.jcmg.2018.11.028. WorldCat 5 Nagy VK , Széplaki G , Apor A , Kutyifa V , Kovács A , Kosztin A et al. Role of right ventricular global longitudinal strain in predicting early and long-term mortality in cardiac resynchronization therapy patients . PLoS One 2015 ; 10 : e0143907. Google Scholar Crossref Search ADS PubMed WorldCat 6 Seo J , Jung IH , Park JH , Kim GS , Lee HY , Byun YS et al. The prognostic value of 2D strain in assessment of the right ventricle in patients with dilated cardiomyopathy . Eur Heart J Cardiovasc Imaging 2019 ; 20 : 1043 – 50 . Google Scholar Crossref Search ADS PubMed WorldCat 7 Lang RM , Badano LP , Mor-Avi V , Afilalo J , Armstrong A , Ernande L et al. Recommendations for cardiac chamber quantification by echocardiography in adults: an update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging . J Am Soc Echocardiogr 2015 ; 28 : 1 – 39.e14 . Google Scholar Crossref Search ADS PubMed WorldCat 8 Kawel-Boehm N , Maceira A , Valsangiacomo-Buechel ER , Vogel-Claussen J , Turkbey EB , Williams R et al. Normal values for cardiovascular magnetic resonance in adults and children . J Cardiovasc Magn Reson 2015 ; 17 : 29 . Google Scholar Crossref Search ADS PubMed WorldCat 9 Schulz-Menger J , Bluemke DA , Bremerich J , Flamm SD , Fogel MA , Friedrich MG et al. Standardized image interpretation and post processing in cardiovascular magnetic resonance: Society for Cardiovascular Magnetic Resonance (SCMR) Board of Trustees Task Force on Standardized Post Processing . J Cardiovasc Magn Reson 2013 ; 15 : 35. Google Scholar Crossref Search ADS PubMed WorldCat 10 Lakatos B , Tősér Z , Tokodi M , Doronina A , Kosztin A , Muraru D et al. Quantification of the relative contribution of the different right ventricular wall motion components to right ventricular ejection fraction: the ReVISION method . Cardiovasc Ultrasound 2017 ; 15 : 8. Google Scholar Crossref Search ADS PubMed WorldCat 11 Maffessanti F , Muraru D , Esposito R , Gripari P , Ermacora D , Santoro C et al. Age-, body size-, and sex-specific reference values for right ventricular volumes and ejection fraction by three-dimensional echocardiography: a multicenter echocardiographic study in 507 healthy volunteers . Circ Cardiovasc Imaging 2013 ; 6 : 700 – 10 . Google Scholar Crossref Search ADS PubMed WorldCat 12 Moceri P , Duchateau N , Baudouy D , Schouver E-D , Leroy S , Squara F et al. Three-dimensional right-ventricular regional deformation and survival in pulmonary hypertension . Eur Heart J Cardiovasc Imaging 2018 ; 19 : 450 – 8 . Google Scholar Crossref Search ADS PubMed WorldCat 13 Nagata Y , Wu V-C , Kado Y , Otani K , Lin F-C , Otsuji Y et al. Prognostic value of right ventricular ejection fraction assessed by transthoracic 3D echocardiography . Circ Cardiovasc Imaging 2017 ; 10 : e005384 . Google Scholar Crossref Search ADS PubMed WorldCat 14 Vitarelli A , Barillà F , Capotosto L , D'Angeli I , Truscelli G , De Maio M et al. Right ventricular function in acute pulmonary embolism: a combined assessment by three-dimensional and speckle-tracking echocardiography . J Am Soc Echocardiogr 2014 ; 27 : 329 – 38 . Google Scholar Crossref Search ADS PubMed WorldCat 15 Crean AM , Maredia N , Ballard G , Menezes R , Wharton G , Forster J et al. 3D Echo systematically underestimates right ventricular volumes compared to cardiovascular magnetic resonance in adult congenital heart disease patients with moderate or severe RV dilatation . J Cardiovasc Magn Reson 2011 ; 13 : 78. Google Scholar Crossref Search ADS PubMed WorldCat 16 Ishizu T , Seo Y , Atsumi A , Tanaka YO , Yamamoto M , Machino-Ohtsuka T et al. Global and regional right ventricular function assessed by novel three-dimensional speckle-tracking echocardiography . J Am Soc Echocardiogr 2017 ; 30 : 1203 – 13 . Google Scholar Crossref Search ADS PubMed WorldCat 17 Muraru D , Spadotto V , Cecchetto A , Romeo G , Aruta P , Ermacora D et al. New speckle-tracking algorithm for right ventricular volume analysis from three-dimensional echocardiographic data sets: validation with cardiac magnetic resonance and comparison with the previous analysis tool . Eur Heart J Cardiovasc Imaging 2016 ; 17 : 1279 – 89 . Google Scholar Crossref Search ADS PubMed WorldCat 18 Park J-B , Lee S-P , Lee J-H , Yoon YE , Park E-A , Kim H-K et al. Quantification of right ventricular volume and function using single-beat three-dimensional echocardiography: a validation study with cardiac magnetic resonance . J Am Soc Echocardiogr 2016 ; 29 : 392 – 401 . Google Scholar Crossref Search ADS PubMed WorldCat 19 Pickett CA , Cheezum MK , Kassop D , Villines TC , Hulten EA. Accuracy of cardiac CT, radionucleotide and invasive ventriculography, two- and three-dimensional echocardiography, and SPECT for left and right ventricular ejection fraction compared with cardiac MRI: a meta-analysis . Eur Heart J Cardiovasc Imaging 2015 ; 16 : 848 – 52 . Google Scholar Crossref Search ADS PubMed WorldCat 20 Sugeng L , Mor-Avi V , Weinert L , Niel J , Ebner C , Steringer-Mascherbauer R et al. Multimodality comparison of quantitative volumetric analysis of the right ventricle . J Am Coll Cardiol Cardiovasc Img 2010 ; 3 : 10 – 18 . Google Scholar Crossref Search ADS WorldCat 21 Genovese D , Rashedi N , Weinert L , Narang A , Addetia K , Patel AR et al. Machine learning-based three-dimensional echocardiographic quantification of right ventricular size and function: validation against cardiac magnetic resonance . J Am Soc Echocardiogr 2019 ; 32 : 969 – 77 . Google Scholar Crossref Search ADS PubMed WorldCat 22 Chuzi S , Rangarajan V , Jafari L , Vaitenas I , Akhter N. Subcostal view-based longitudinal strain in patients with breast cancer is an alternative to conventional apical view-based longitudinal strain . J Am Soc Echocardiogr 2019 ; 32 : 514 – 20 . Google Scholar Crossref Search ADS PubMed WorldCat Published on behalf of the European Society of Cardiology. All rights reserved. © The Author(s) 2019. For permissions, please email: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - Accuracy of fully automated right ventricular quantification software with 3D echocardiography: direct comparison with cardiac magnetic resonance and semi-automated quantification software
JF - European Heart Journal - Cardiovascular Imaging
DO - 10.1093/ehjci/jez236
DA - 2020-07-01
UR - https://www.deepdyve.com/lp/oxford-university-press/accuracy-of-fully-automated-right-ventricular-quantification-software-xdnudKoqtx
SP - 1
VL - Advance Article
IS - 
DP - DeepDyve
ER -