Automated determination of the left ventricular long axis in cardiac positron tomography

Interpretation of emission tomographic images of the heart is typically performed using short-axis sections which are oriented perpendicular to the long axis of the left ventricle. A completely automated method is presented to find the orientation and length of the long axis in positron emission tomographic studies of the heart. The correlation coefficient is maximized between the measured transaxial images and an ellipsoid model of the left ventricular myocardium. The major axis of this fitted ellipsoid corresponds to the long axis of the left ventricle. The orientation and position of the long axis (with respect to the measured transaxial images) define two rotation angles and a 3D coordinate origin which are used to re-orient the transaxial images into a series of standard short-axis sections. The accuracy and precision of this technique are validated on phantom data as well as on patients with documented myocardial infarction. The transaxial and vertical long-axis rotation angles are determined with standard deviations of 3.3 degrees and 1.5 degrees, respectively. The 3D coordinate origin (centre) of the ellipsoid is accurate to within 1.5 mm on average. The estimated length of the left ventricle is accurate to within 3 mm. All parameters are insensitive to statistical noise found in typical 18F-fluorodeoxyglucose (FDG) patient studies. The technique produces accurate estimates even in the presence of moderate uptake defects in the lateral wall and septum. Apical defects in the uptake of FDG do not increase the variance of the length estimate. These results demonstrate that an ellipsoid model can be fitted accurately to the myocardium of the left ventricle. Standard short-axis sections are produced with no inter-operator or intra-operator variability because the technique is fully automatic.