Background
Recently, a new automated software package (HeartModel) was developed to obtain three-dimensional (3D) left ventricular (LV) volumes using a model-based algorithm (MBA) with a “one-button” simple system and user-adjustable slider. The aims of this study were to verify the feasibility and accuracy of the MBA in comparison with other commonly used imaging techniques in a large unselected population, to evaluate possible accuracy improvements of free operator border adjustments or changes of the slider’s default position, and to identify differences in method accuracy related to specific pathologies.
Methods
This prospective study included consecutive 200 patients. LV volumes and ejection fraction were obtained using the MBA and compared with the two-dimensional biplane method, the 3D full-volume (3DFV) modality, and, in 90 of 200 cases, cardiac magnetic resonance (CMR) measurements. To evaluate the optimal position of the slider with respect to the 3DFV and CMR modalities, a set of threefold cross-validation experiments was performed. Optimized and manually corrected LV volumes obtained using the MBA were also tested. Linear correlation and Bland-Altman analysis were used to assess intertechnique agreement.
Results
Automatic volumes were feasible in 194 patients (94.5%), with a mean processing time of 29 ± 10 sec. MBA-derived volumes correlated significantly with all evaluated methods, with slight overestimation of two-dimensional biplane and slight underestimation of CMR measurements. Higher correlations were found between MBA and 3DFV measurements, with negligible differences both in volumes (overestimation) and in LV ejection fraction (underestimation), respectively. Optimization of the user-adjustable slider position improved the correlation and markedly reduced the bias between the MBA and 3DFV or CMR. The accuracy of MBA volumes was lower in some pathologies for incorrect definition of LV endocardium.
Conclusions
The MBA is highly feasible, reproducible, and rapid, and it correlates highly with the traditional 3DFV method. It may represent a valid alternative to 3DFV measurement for everyday clinical use.
Highlights
- •
The authors analyzed the feasibility and accuracy of a new echocardiographic 3D automatic MBA for LV volume and functional evaluation.
- •
The new automatic method was applied in a large population and enabled LV volume and LVEF measurements in a few seconds.
- •
The authors tested and found the best setting of the slider position for the automatic definition of LV wall borders, optimizing correlation between the new method and 3D traditional full-volume or CMR.
An accurate and reproducible assessment of left ventricular (LV) volumes and function is very important in all cardiac diseases and is the most frequent indication for an echocardiographic study in daily practice. The geometric assumptions necessary to obtain LV volumetric reconstruction and the suboptimal inter- and intraobserver variability in two-dimensional (2D) echocardiography are known to limit this technique. The introduction of three-dimensional (3D) echocardiography with LV dedicated software has allowed a more reliable analysis of LV volumetric and functional data, thus increasing reproducibility in comparison with 2D echocardiography and accuracy in comparison with cardiac magnetic resonance (CMR).
However, especially at the beginning, cumbersome acquisition methods and complicated and time-consuming analysis software reduced the diffusion of 3D echocardiography in routine LV evaluation. Improvements in “on-board” semiautomatic volumetric methods have allowed increasing use of LV 3D echocardiographic quantification, even though, for everyday clinical use, 3D echocardiographic LV volumetric evaluation will be ready only with the introduction of more simple and fast acquisition modalities and automatic chamber quantification techniques.
Recently, a new automated software package has been developed to obtain LV volumes from real-time 3D echocardiographic acquisitions using a model-based adaptive analytic algorithm with a “one-button” simple system and a user-adjustable slider for the detection of endocardial borders. This new method was recently evaluated, demonstrating that simultaneous quantification of left atrial and LV volumes and LV ejection fraction (LVEF) is feasible and requires minimal 3D software analysis training.
The aim of this study was threefold: (1) to verify the feasibility and accuracy of the model-based algorithm (MBA) in comparison with the 2D biplane (2DBP) method, 3D full-volume (3DFV) modality (3DFV), and CMR in a large population of patients; (2) to evaluate if changes in MBA volumetric reconstruction through operator border adjustments or repositioning of the slider border definition might improve the MBA’s accuracy; and (3) to identify differences in method accuracy relating to specific pathologies.
Methods
We prospectively recruited 200 consecutive patients in sinus rhythm referred to the echocardiography laboratory of Centro Cardiologico Monzino of Milan for measurement of LV volumes and LVEF. After the exclusion of six patients with technically inadequate echocardiographic images, a study group of 194 patients remained who underwent 2D and 3D echocardiography.
For clinical reasons, CMR studies were performed in 90 of these 194 patients. The population consisted of 34 normal subjects and 160 patients with valve disease ( n = 68), coronary artery disease ( n = 23), dilated cardiomyopathy ( n = 53), and congenital or hypertrophic disease ( n = 16). The study protocol conformed to the ethical guidelines of the 1975 Declaration of Helsinki as reflected in a priori approval by the institution’s human research committee and was approved by the institutional review board. Informed consent was obtained from each patient.
2D Echocardiography
All echocardiographic examinations were performed using a Philips echocardiographic system (EPIQ, iE33, X5 transducer; Philips Healthcare, Andover, MA). A complete standard 2D echocardiographic examination was performed. Biplane LV end-diastolic volume (LVEDV), LV end-systolic volume (LVESV), and LVEF were measured from the four- and two-chamber views using Simpson’s method.
3D Echocardiography
With the same echocardiographic system and same transducer, at the end of the two-dimensional echocardiographic study, 3D echocardiographic acquisitions were obtained from the four-chamber apical view in full-volume mode (3DFV), gathered over four cardiac cycles, during a breath-hold lasting a few seconds.
3DFV
For the semiautomatically derived 3D echocardiographic method, LV volumetric data sets were measured using commercially available software (QLAB-3DQ Adv; Philips Healthcare). Briefly, the operator aligns the multiplanar view to optimize horizontal and vertical lines in the middle of the LV cavity. Then five reference points are placed (septal, lateral, anterior, and inferior mitral annulus and apex) at the end-diastolic and end-systolic frames. Both the end-diastolic and end-systolic frames are automatically obtained by the software, but if necessary, the correct frame may be modified by the operator. Finally, the software automatically identifies the LV endocardial border and creates a 3D LV model providing LV volumes and calculating LVEF. A suboptimal automatic endocardial border delineation may be manually adjusted when necessary.
Model-Based Algorithm
This new 3D echocardiographic software involves an automated analysis that simultaneously detects LV and left atrial endocardial surfaces using an adaptive analytics algorithm. The program and methodology of the system have been previously described. In brief, in each patient two or more acquisitions were performed using the new automatic method from the same four-chamber apical window during a brief breath-hold period. When LV acquisition was obtained, touching the icon of the MBA software on the echocardiography screen, LV volumes, stroke, and LVEF were calculated without operator intervention ( Figure 1 ). MBA software automatically detects the LV wall inner border at the blood-tissue interface and outer border located at compacted myocardium interface. In the default setting, a slider is positioned in the middle between the two borders (default setting = 50), and in this position LV volume is automatically assembled. However, this user=adjustable slider may be freely moved from the default position to arbitrarily optimize LV border identification, and different slider positions can be preset to a user’s preference. The default slider position was used as the MBA reference value both for end-diastolic and end-systolic volumes. Figure 1 shows an example of MBA imaging.
Different User-Adjustable Slider Positions and Free Adjustment of the Automatic Border
To improve the correlations between MBA and CMR or 3DFV, two correction modalities of automatic MBA reconstruction were used: fixed changes of the default slider position and free adjustment of the automatic border.
For fixed changes of default slider position, the operator moves the user-adjustable sliders toward the optimal blood-tissue interface or outer border at compacted myocardium interface to optimize the endocardial recognition (MBA optimized). For free adjustment of the automatic border the operator adjust a specific region of the endocardial border (MBA free). Thus, we analyzed data obtained using the automatic default slider position (MBA), MBA optimized, and MBA free.
To evaluate optimal position of the slider with respect to 3DFV and CMR modalities, when adjusting the LV borders computed with MBA optimized, a set of threefold cross-validation experiments was performed. We set the number of folds equal to three because it enabled us to achieve a good bias-variance trade-off. For 3DFV, 126 patients were used to train the position of the user-adjustable slider, and the remaining 63 were used for validation.
For CMR, the position of the slider was trained on end-diastolic and end-systolic volumes of 56 patients and then tested on the remaining 28. Finally, correlations were performed between MBA-adjusted volumes and 3DFV and CMR measurements respectively.
In all cases, free adjustment of the automatic border detection was also performed by the operator with changes of the slider position between the inner and outer borders and/or with manually regional corrections to arbitrarily optimize endocardial identification. The MBA free-adjusted volumes were then compared both with the corresponding 3DFV and CMR measurements.
Feasibility and Duration of the Method
Feasibility was evaluated as the percentage of patients in whom MBA LV volumes were obtained in comparison with all patients enrolled in the study population. Adequate MBA reconstruction was considered when endocardial border of the LV cavity was correctly delineated and ventricular-atrial annulus recognized in the correct end-systolic and end-diastolic frames.
LV volume acquisition and reconstruction time with the different echocardiographic modality (2DBP, 3DFV, and MBA) were annotated in each case.
CMR Protocol
For specific clinical reasons and diagnostic indications, CMR was performed in 90 patients using a 1.5-T scanner (Discovery MR450; GE Healthcare, Milwaukee, WI) within 2 days of the echocardiographic examination. After the acquisition of localizer images of the heart, breath-hold steady-state free precession cine acquisitions were acquired using the following parameters: echo time 1.57 msec, 15 segments, repetition time 46 msec without view sharing, slice thickness 8 mm, field of view 350 × 263 mm, and pixel size 1.4 × 2.2 mm. CMR data were transferred to a dedicated workstation and analyzed using dedicated cardiac software (Report Card 4.0; GE Healthcare). LVEDV, LVESV, and LVEF were evaluated on cine images according to the recommendations of the Society of Cardiovascular Magnetic Resonance.
Statistical Analysis
Results for LVEDV, LVESV, and LVEF are presented as mean ± SD.
LVEDV, LVESV, and LVEF obtained with the MBA were compared with 2DBP, 3DFV, and CMR values using linear correlation and Bland-Altman analysis. A P value of <.05 was considered to indicate statistical significance.
Data for test-retest reliability of the MBA were obtained by the same operator (G.T.) in 100 patients by removing the probe after the first acquisition and repositioning the transducer after 5 min to obtain the second data set. MBA volumes were automatically obtain without operator intervention.
An independent investigator (M.Mu.) who was blinded to the MBA results performed 3DFV evaluation on a separate day in all patients. To assess the reproducibility of the 3DFV measurements in a subset of 50 randomly chosen subjects, M.Mu. reevaluated the same 3D data sets 2 weeks after the first analysis, blinded with respect to the results of the previous evaluation. For each computed parameter, intraobserver variability was then evaluated. The same subset was also evaluated by a third different observer (M.P.), blinded to the results obtained by M.Mu., to assess interobserver variability.
Intra-class correlation coefficient (ICCs) and coefficients of variation (100 × SD/mean) on computed volumes were used to assess variability. Good reproducibility was indicated by an ICC > 0.75 or a coefficient of variation < 20% between measurements.
All investigators who performed echocardiographic analyses were blinded to the CMR results.
Results
Table 1 shows the clinical characteristic of the study patients. At least one good MBA reconstruction was feasible in 189 of 200 cases. After the exclusion of six patients with technically inadequate echocardiographic images, a study group of 194 patients underwent 2DE and 3DE. However despite satisfactory acquisition, in three patients, MBA reconstruction was not obtained, because of incorrect recognition of LV ventricular apex or ventricular-atrial junction. In two cases, the MBA was not evaluable, because of incorrect automatic identification of the end-systolic frame. In these five cases, 3DFV LV volumes were calculated because of the possibility of manual frame adjustments and/or manual border tracing. Therefore, our study population was composed of 189 cases.
| Variable | Total | With CMR |
|---|---|---|
| Number of patients | 189 | 90 |
| Women | 59 | 26 |
| Age (y) | 59.8 ± 15 | 59.4 ± 15 |
| BSA (m 2 ) | 1.8 ± 0.2 | 1.8 ± 0.2 |
| % MBA feasibility | 94.5% | 100% |
The mean time for acquisition and automatic volume reconstruction was 29 ± 10 sec. The time needed (acquisition and analysis) for LV volume and LVEF estimation was longer with the 3DFV (160 ± 30 sec) and 2DBP (120 ± 20 sec) methods.
Test-retest analysis of MBA measurement showed that sequential volume measurements correlated significantly (LVEDV, r = 0.98; LVESV, r = 0.99), with limited bias (LVEDV, 3.1 mL; LVESV, 0.5 mL).
Intraobserver variability of 3DFV measurements of LVEDV and LVESV evaluated with coefficients of variation were 3.38% and 6.05%, respectively, while interobserver variability of the same measures was 5% and 10.72%, respectively. When evaluating the variability of the same clinical parameters with ICCs, high values of reproducibility were observed (intraobserver variability: ICC = 0.9975 for LVEDV, ICC = 0.9974 for LVESV; interobserver variability: ICC = 0.9826 for LVEDV, ICC = 0.9734 for LVESV).
Automatic Measurements versus 2DBP, 3DFV, and CMR
Table 2 shows the results of the comparisons between the 2D echocardiographic, 3D echocardiographic, and CMR measurements of LVEDV, LVESV, and LVEF against the corresponding MBA values.
| Variable | MBA | Other technique | Correlation | Bias | Limits of agreement |
|---|---|---|---|---|---|
| 2DBP measurements (189 patients) | |||||
| LVEDV (mL) | 175.1 ± 64 | 142.4 ± 58 | 0.90 | −32.8 | 56.4 |
| LVESV (mL) | 102 ± 59 | 71.0 ± 49 | 0.92 | −30.8 | 46.9 |
| LVEF (%) | 44.5 ± 14 | 53.8 ± 15 | 0.80 | 9.2 | 17.3 |
| 3DFV measurements (189 patients) | |||||
| LVEDV (mL) | 175.1 ± 64 | 165.1 ± 63 | 0.97 | −9.5 | 29.9 |
| LVESV (mL) | 102 ± 59 | 85.5 ± 56 | 0.97 | −16.4 | 27.9 |
| LVEF (%) | 44.5 ± 14 | 51.6 ± 14 | 0.88 | 7.3 | 12.9 |
| CMR (84 patients) | |||||
| LVEDV (mL) | 175.2 ± 61 | 200.1 ± 74 | 0.90 | 25.7 | 65.8 |
| LVESV (mL) | 107.6 ± 59 | 117.4 ± 73 | 0.91 | 9.6 | 62 |
| LVEF (%) | 41.6 ± 13 | 45.3 ± 17 | 0.79 | 4 | 20 |
MBA-derived values correlated significantly with 2DBP measurements despite an important overestimation both of diastolic and systolic LV volumes and an underestimation of LVEF. The corresponding 3DFV measurements resulted in a higher correlations with MBA, with negligible differences both in volumes (overestimation) and in LVEF (underestimation).
A significant correlation was also present between MBA and CMR values, with a reasonable bias for LV volumes due to MBA underestimation and a very limited bias for LVEF but with large limits of agreement.
MBA and 3DFV measurements were separately compared in normal and in each pathologic group. As shown in Table 3 , the largest differences were observed for congenital and hypertrophic diseases, with an important MBA overestimation in particular of LVESV.
