Automated 3D elastic registration for improving tumor localization in whole-body PET-CT from combined scanner

Combined PET/CT scanners provide the ability to produce matching metabolic (from PET) and anatomic (from CT) information in a single examination. However, misalignments continue to exist in tumor localization in PET and CT images acquired using these scanners, due to their inability to compensate for nonrigid misalignment resulting from patient breathing and involuntary movement. We demonstrate that our automatic image subdivision-based elastic registration algorithm can correct this misalignment. In a quantitative validation involving 13 expert-identified tumor nodules in six PET-CT image pairs, the algorithm demonstrated statistically significant improvement over the scanner-defined localization. The accuracy of algorithm-determined localization was evaluated to be comparable to average manually defined localization. The results indicate the potential of using our registration algorithm for applications like radiotherapy treatment planning and treatment-monitoring involving combined PET/CT scanners.