Search

The purpose of the present study was to evaluate the robustness of a method of automated border detection in cardiac magnetic resonance (MR) imaging. Thirty-seven short-axis spin-echo cardiac images were acquired from three medical centers, each with its own image-acquisition protocol. Endo- and epicardial borders and areas were derived from these images with a graph-searching-based method of edge detection. Computer results were compared with observer-traced borders. The method accurately defined myocardial borders in 36 of 37 images (97%), with excellent agreement between computer- and observer-derived endocardial and epicardial areas (correlation coefficients, .94-.99). The algorithm worked equally well for data from all three centers, despite differences in image-acquisition protocols, MR systems, and field strengths. These data suggest that a method of computer-assisted edge detection based on graph-searching principles yields endocardial and epicardial areas that correlate well with those derived by an independent observer.

Gated cardiac magnetic resonance imaging (MRI) permits detailed evaluation of cardiac anatomy, including the calculation of left ventricular volume and mass. Current methods of deriving this information, however, require manual tracing of boundaries in several images; such manual methods are tedious, time consuming, and subjective. The purpose of this study is to apply a new computerized method to automatically identify endocardial and epicardial borders in MRIs. The authors obtained serial, short-axis, spin-echo MRIs of 13 excised animal hearts. Also obtained were selected short-axis, spin-echo ventricular images of 11 normal human volunteers. A method of automated edge detection based on graph-searching principles was applied to the ex vivo and in vivo images. Endocardial and epicardial areas were used to compute left ventricular mass and were compared with the anatomic left ventricular mass for the images of excised hearts. The endocardial and epicardial areas calculated from computer-derived borders were compared with areas from observer tracing. There was very close correspondence between computer-derived and observer tracings for excised hearts (r = 0.97 for endocardium, r = 0.99 for epicardium) and in vivo scans (r = 0.92 for endocardium, r = 0.90 for epicardium). There also was a close correspondence between computer-generated and actual left ventricular mass in the excised hearts (r = 0.99). These data suggest the feasibility of automated edge detection in MRIs. Although further validation is needed, this method may prove useful in clinical MRI.