Example of apical 4- and 2-chamber (top left and right, respectively) and short-axis (bottom left) cut planes obtained from a real-time 3-dimensional echocardiographic (RT3DE) dataset of 1 patient. Images are shown with semiautomatically traced endocardial contours that include endocardial trabeculae in the left ventricular (LV) cavity. Optimization of the boundaries in multiple planes results in a cast of the LV cavity (bottom right), from which LV volume is quantified by counting voxels inside the cast.

This is demonstrated here using a long-axis CT image (left) depicting left ventricular (LV) anatomy with great detail. Cardiac magnetic resonance images are generated by compacting information from slices of finite thickness (8 mm in this study). A slice that contains the mitral valve (middle panel, between horizontal orange lines) would correspond to a short-axis view where most of the LV cavity circumference is surrounded by myocardial tissue (right). By convention, this slice would be included in the calculation of LV volume. Since this rule is often difficult to use with confidence because of partial volume artifacts, it may affect CMR-derived LV volumes. We sought to determine to what extent the volume in question, V (middle panel), would effect LV volume measurements in a group of consecutive patients.

To obtain CMR reference values without analysis-related intertechnique differences, CMR images were reformatted into 3D datasets and analyzed using the same software that was used to analyze RT3DE data. Apical 4- and 2-chamber (top left and right, respectively) and short-axis (bottom left) cut planes as well as the LV cast (bottom right) obtained from the 3D CMR dataset of the same patient shown in the same format as in Figure 1. Abbreviations as in Figures 1 and 2.

These examples of short-axis cut planes extracted from RT3DE datasets demonstrate how spatial resolution may affect the perception of endocardial boundaries. In 1 patient (left), endocardial trabeculae can be well visualized and clearly differentiated from the myocardium and thus appropriately included in the LV cavity. In contrast, in the second patient (right), the spatial resolution of the RT3DE image is not sufficient to provide this kind of detail and is likely to result in erroneous exclusion of the trabeculae from the LV cavity. Abbreviations as in Figure 1.

These plots show the results of linear regression and Bland-Altman analyses between RT3DE-derived LV volumes (end-diastolic volume [EDV] and end-systolic volume [ESV]) and ejection fraction (EF) against CMR reference values obtained in 92 study patients: r values and difference from CMR values averaged over patients (bias) ± standard deviation (95% limits of agreement [LOA]). Despite the high correlations that were similar to previously published single-center studies, RT3DE-derived volumes showed large negative biases in the multicenter setting. Abbreviations as in Figures 1 and 2.

(Top) Long-axis cut plane extracted form a RT3DE dataset of an egg-shaped phantom, shown with the boundary traced along the interface (A), after expanding the boundary 1 mm outward (B), and with both boundaries (C). Interestingly, the small difference between the 2 boundaries resulted in an 11% difference in the measured volume of the 3D shell (see text for details). (Bottom) Cross-sectional views of a water-filled latex balloon with 3 alternative manually traced boundaries: along the inner interface (D), along the outer interface (E), and in the center of the latex layer (F). Volumes (V) resulting from each tracing session are shown to be compared with the true volume of 150 ml. Abbreviations as in Figure 1.

Results of linear regression and Bland-Altman analyses of agreement between LV volumes measured in 19 patients using volumetric analysis of interpolated 3D CMR datasets (QLAB Advanced software) and the conventional method of disk approximation (ViewForum software, Philips). The high levels of agreement evidenced by these results ruled out analysis-related differences between RT3DE and CMR measurements as a significant source of error. LOA = limits of agreement; other abbreviations as in Figures 1, 2, and 5.

(Left) Example of a short-axis CMR slice extracted from an interpolated 3D CMR dataset with endocardial surface traced to include trabeculae (A) and, in a separate analysis, to exclude them (B). This experiment was performed with data obtained in 19 patients. (Right) Results of linear regression and Bland-Altman analyses are shown (C: EDV, D: ESV) for both sets of CMR reference values, those obtained with (yellow) and without (orange) the trabeculae as part of the LV cavity (see text for details). Abbreviations as in Figures 1, 2, and 5.