Fully automatic segmentation of multiaxis CMR images produced comparable results to manual segmentation for left ventricular volume, but required 2 minutes of user correction for the right ventricle.
Observational (n=40)
Yes
Does automatic biventricular segmentation perform comparably to manual segmentation in multiaxis CMR images?
Automatic 3D surface model segmentation of multiaxis CMR images is accurate for the left ventricle but requires minor manual correction for the right ventricle.
PURPOSE: To validate a volumetric biventricular segmentation solution for multiaxis cardiac magnetic resonance (CMR) images. MATERIALS AND METHODS: The study population comprised 40 subjects. Biventricular end-diastolic and -systolic phases were segmented from both short-axis and horizontal long-axis or transaxial cine CMR images. Segmentation was based on fitting nonrigidly a 3D surface model to multiaxis CMR images. Five segmentations were performed: two manual segmentations by experts, automatic segmentation, and two segmentations where a user was allowed to correct errors in the automatic segmentation for 2 minutes and without time limits. Volumetry, distance measures, and visual grading were used to evaluate the quality of the segmentation. RESULTS: No difference was observed between automatic and manual segmentations in volumetric measures of the ventricles. The manual segmentation performed better for left-ventricular myocardial volume. The distance between surfaces as well as visual analysis did not show differences between automatic and manual segmentation for the endocardial border of the left ventricle but some corrections are needed for the right ventricle. CONCLUSION: Fully automatic segmentation produces good results in the assessment of left ventricular volume andendocardial border. Two minutes of user interaction are needed to obtain accurate results for the right ventricle.
Lötjönen et al. (2008) reported an observational. Automatic volumetric biventricular segmentation vs. Manual segmentation was evaluated on Volumetry, distance measures, and visual grading. Fully automatic segmentation of multiaxis CMR images produced comparable results to manual segmentation for left ventricular volume, but required 2 minutes of user correction for the right ventricle.