A. Bazille et al., IMPACT OF SEMIAUTOMATED VERSUS MANUAL IMAGE SEGMENTATION ERRORS ON MYOCARDIAL STRAIN CALCULATION BY MAGNETIC-RESONANCE TAGGING, Investigative radiology, 29(4), 1994, pp. 427-433
RATIONALE AND OBJECTIVES. Accurate image segmentation is essential for
the study of mechanical properties of the myocardium by tagged magnet
ic resonance imaging (MRI). The relative accuracy of three methods of
segmentation of myocardial borders and tags and their impact on myocar
dial strain calculations were evaluated. METHODS. Radially tagged, spi
n-echo magnetic resonance (MR) images of dog hearts were segmented man
ually, automatically, and semiautomatically. The variability of segmen
tation methods was separately determined for myocardial contours and t
ags. Error propagation assessment for strain calculation was estimated
. RESULTS. The variability of the segmentation of the contours was fiv
e times greater than that of the tags. The error propagation is nonuni
form in all directions, maximal for the radial component of strain. CO
NCLUSION. Errors in the segmentation of myocardial contours are signif
icantly greater than those of the tags. Strain calculations should be
based solely on segmentation of MR tags to avoid significant errors, p
articularly in radial strain estimates.