Use of spline interpolation for border definition from two-dimensional echocardiograms: comparison to manual method

A semiautomatic method for definition of cardiac chamber borders from two-dimensional echocardiograms has been developed which incorporates spline interplation to produce a smooth cure with moderate operator involvement. This research compares the semiautomatic border definition (SBD) method to the currently used manual border definition (MBD) method according to the following criteria: time requiarements, storage requirements, reproducibility, and accuracy. A paired t-test showed the mean contour generation time over multiple observers, patients, trials and cardiac contours to be significantly greater for SBD than for MBD (p=0). The mean number of stored border points for SBD was 9 ± 3.84. Manual contours each required 64 stored border points to maintain contour integrity. Reproducibility of 6 cardiac contours was compared via a sign test analysis of variances of 64 radial or hemichordal lengths recorded from contours generated over multiple observers, trials and patients. SBD produced significantly more reproducible contours for the short axis view at the mitral valve level at end-diastole (p-0) and at end-systole (p=0) and for the apical four-chamber view at end-diastole (p=0). There was no significant difference in reproducibility between methods for the apical four-chamber view at end-systole (p=.136) and the apical two-chamber view at end-diastole (p=.784) Only the apical two-chamber view at end-systole proved to be significantly more reproducible using MBD (p=.046). Accuracy of methods was compared by generating contours from black-on-white phantom images for which the true border coordinates were known. A mean absolute radial difference was calculated from 64 radial measurements taken from each contour and a paired t-test comparing these values between methods showed SBD to be significantly more accurate then MBD (p=0). Time requirements for SBD are highly machine dependent and may be significantly reduced with upgraded hardware. Subjectively, SBD appears to be less tedious and less fatiguing for the operator. For these reasons and because it has been shown more accurate and equally reproducible or more reproducible contours than MBD, SBD appears to be the preferred method for two-dimensional echocardiographic contour generation.