Session M1.5
A Comparison of 2D and 3D Edge Detectors in Semi-Automated Measurements of Chamber Volumes Using 3D Echocardiographic Laboratory Phantom Images
K Wang*, AJ Sims, A Murray
Newcastle University
Newcastle upon Tyne, UK
Cardiac chamber wall delineation and volume measurement are fundamental for assessing a wide range of indices of cardiac function. Three-dimensional echocardiography offers great potential for better assessing heart chamber functions at the cost of increased effort of data handling. Automatic or semi-automatic image analysis techniques developed recently can help to reduce the manual work, but tend to suffer from effects of acoustic noise and artefacts. Suitable choice of edge detection operators can reduce the effect of noise. The aim of this research is to compare the performance of 2D and 3D first order derivative edge detection operators (11×11 pixels, 5×5×5 pixels) in six semi-automatic quantitative measurement methods by volume assessment of 3D echocardiographic images of a laboratory balloon phantom. Nine fill volumes (range from 25ml to 500ml) of the balloon phantom were scanned by 3DE. For the 3D image of each volume, five measurements were performed using each semi-automatic method and each operator. We applied the Bland-Altman method to compare the measured volumes of 2D and 3D operators with known volumes of the balloon. The results showed that overall, the measured volumes using the 2D operator were significantly different from the known volumes (-11.6ml; -14.8ml to -8.5ml 95% CI; P<0.001), but the measured volume of the 3D operator was not (-1.0ml; -4.0ml to 2.0ml 95% CI; P=0.53). We also compared the coefficient of variation (CV) and standard deviation (SD) of two operators. A paired student t test showed that the CV with the 3D operator was significantly smaller than with the 2D operator (-0.6; P<0.001). Analysis of variance also indicated that the SD with the 3D operator was significantly smaller than with the 2D operator (F=11.0; P=0.001; 2D: mean SD=1.5 ml; 3D: mean SD = 0.9 ml). In conclusion, compared with the 2D first order differential gradient operator, the 3D operator can improve the accuracy as well as the reproducibility of quantitative volume measurement in real time 3D echocardiographic images when both gradient operators contain a similar number of pixels.
(Abstract Control Number: 196)