Feedback-assisted three-dimensional reconstruction of the left ventricle with MRI

PURPOSE: To develop and test a new technique for rapid, accurate three-dimensional (3D) reconstruction of the left ventricle (LV) and calculation of its volume parameters, with images from multiple orientations and interactive feedback.
MATERIALS AND METHODS: The ventricular surface was fit to a number of user-placed guide points in magnetic resonance (MR) images using bivariate smoothing splines. A 3D model was reconstructed and the LV volumes were calculated at both end diastole (ED) and end systole (ES). This technique was validated using a phantom, and applied to studies of 18 patients and four volunteers (N = 22) imaged on a 1.5-T clinical scanner. The results of the 3D method were compared to the standard 2D short-axis slice summation technique, which is widely used for the analysis of cardiac function.
RESULTS: There was excellent agreement between the computed volume of the phantom using the 3D modeling method and the actual volume (190.50 mL +/- 3.06 mL, and 191.0 mL +/- 2.5 mL, respectively). There was good correlation between the volumes calculated with our 3D model and the slice summation technique (ED volume (EDV) difference, 6.36% +/- 8.99% [mean +/- SD]; ES volume (ESV), 0.92% +/- 14.75%; stroke volume (SV), 10.54% +/- 13.95%; ejection fraction (EF), 4.22% +/- 9.16%). The 3D method was found to be more accurate than the slice summation technique for calculating LV volumes and mass from images of different slice orientations. Variations in the parameters between the two separate orientations using the 3D model vs. the slice summation method were as follows: EDV: 2.11% +/- 1.52% vs. 10.36% +/- 9.33%; ES volume: 2.76% +/- 1.64% vs. 6.39% +/- 3.62%; SV, 3.02% +/- 4.38% vs. 18.84% +/- 15.30%; EF, 2.03% +/- 2.16% vs. 8.58% +/- 6.73%; and LV mass: 4.77% +/- 2.41% vs. 24.59% +/- 6.41%. Differences in the ES volume due to the inclusion or exclusion of the most basal slice were found to be lower with the 3D model (6.90% +/- 3.83%) compared to the slice summation method (25.04% +/- 6.15%).
CONCLUSION: 3D models can be used to accurately determine ventricular volume parameters. Results can be obtained using images from a variety of orientations, providing greater flexibility during image acquisition and possibly reducing the number of images needed for analysis. Feedback is provided to assist the analysis by providing a continuous update of the LV shape and volume. This feature allows the user to determine LV parameters to a predefined accuracy or to terminate the analysis when the parameters are not changing. This method is not restricted to multislice cine imaging in a single or prescribed slice orientation, and can be used for quick, accurate, and interactive analysis of cardiac function.