Purpose: The aim of the study was to determine the accuracy of non-rigid nine-parameter image registrations based on 153Gd transmission computed tomography (TCT) images as compared with those based on 99mTc-ethyl cysteinate dimer (ECD) images and to assess whether normalised mutual information (NMI) or count difference (CD) should be used.
Methods: TCT and ECD data were acquired in 25 randomly selected patients. Emission images were registered to an ECD template with a CD cost function. The same registration parameters were applied to the transmission images to create a TCT template. All TCT images were registered to the TCT template and the same registration parameters were applied to the ECD images. The procedure was repeated with NMI as cost function. Accuracy of both ECD-based and TCT-based registrations was assessed by comparing the normalisation parameter values and regional activities in the spatially normalised ECD images, using a mixed-model analysis of variance (ANOVA). Scheffe post hoc tests were performed.
Results: No significant differences were found between ECD/CD, ECD/NMI and TCT/CD, suggesting that ECD registration can be done with either CD or NMI, and that TCT registration using CD is equally as accurate as ECD registration. The accuracy of TCT registration with NMI was lower, with discrepancies occurring in the frontal inferior region and the cerebellum. The analysis of normalisation parameters indicated that z-scaling is underestimated and yz-rotation overestimated with TCT/NMI registration.
Conclusion: We conclude that ECD registrations with CD or NMI are as accurate as TCT registrations with CD and that TCT registrations with NMI should be avoided.