In myocardial SPECT perfusion imaging, reorientation algorithms from transaxial image planes are used to generate short- and long-axis views of myocardial tracer uptake. We performed phantom experiments with 201Tl to delineate how image reorientation affects the results of quantitative image analysis.
Methods: Thirty consecutive patient studies were analyzed to characterize the distribution of the angle of reorientation in a clinical setting. Short-axis SPECT images of a cardiac phantom with and without a 180 degrees cold-spot insert were reconstructed with three different backprojection filters (ramp, Metz and Butterworth) and reoriented through different angles ranging from 45 degrees to 89 degrees. Four interpolation algorithms were used to calculate from the transaxial images the pixel values of the reoriented images: (a) a simple interpolator that averages the pixel values of the eight neighboring pixels of the transaxial image; (b) a three-dimensional linear interpolator; (c) a hybrid interpolator that combines a two-dimensional linear in-plane with a one-dimensional cubic across-plane interpolation; and (d) a three-dimensional cubic convolution interpolator. Images were reoriented twice with opposite angles so that the original and the reoriented images could be directly compared. Circumferential profile analysis was applied to determine the root mean square error of corresponding profiles and the difference of the extent and the severity of perfusion defects. Single and multivariate analyses of variance (ANOVA) were used to compare the effects of the reorientation angle, the backprojection filter and the interpolation algorithm.
Results: In the clinical studies, the angle between the transaxial and reoriented images was 75 degrees +/- 10 degrees (s.d.). In 48 phantom experiments, multivariate ANOVA demonstrated that the backprojection filter and the interpolation algorithm significantly affect the circumferential profiles and the extent and severity of a perfusion defect (p < 0.05). In contrast, the angle of reorientation was not a significant factor (p = ns). By univariate analysis, the three-dimensional cubic interpolator was associated with significantly (p < 0.05) less error than the simple and three-dimensional linear algorithms. Relative computation times (simple interpolator = 100%) were 119% for the three-dimensional linear, 136% for the hybrid and 243% for the three-dimensional cubic interpolator.
Conclusion: For quantitative analysis of myocardial SPECT perfusion images, a Metz filter for filtered backprojection in combination with a three-dimensional cubic convolution interpolation for image reorientation appears to offer improved accuracy.