Comparison of performance between rigid and non-rigid software registering CT to FDG-PET

Object: This retrospective study compares the anatomical accuracy of automated rigid and non-rigid registration software for aligning data from separately performed X-ray computed tomography (CT) and positron emission tomography with F-18-deoxyglucose (PET). Materials and methods: Analyses were performed on independently acquired PET and CT data from 40 tumor patients. Rigid as well as non-rigid automated fusion was carried out using the commercially available Mirada 7D platform (MIR and MINR, respectively) as well as a second automated non-rigid registration based on a variational image registration approach (VIR). Distances between lesion representation on PET and CT of 105 malignant lesions were measured in X-, Y-, and Z-directions. Statistical evaluation was performed using mixed effect analysis, comparing separately MIR with MINR and VIR with MINR. Results: The percentage of lesions misregistered by less than 15 mm varied from 70% for MIR and MINR in Z-direction to 93% for VIR in X-direction. The average X-, Y- and Z-distances ranged between 5.9 ± 5.7 mm for VIR in X-direction and 12.8±9.7 mm for MIR in Z-direction. MINR was significantly more accurate than MIR in Y-direction. Furthermore, VIR aligned thoracic lesions in the X- direction significantly better than MINR. Conclusion: The accuracy of rigid and non-rigid automated image registration can be expected to be better than 15 mm for the majority of lesions. Alignment tended to be more accurate with non-rigid registration. This work was supported by the ELAN-Fonds of the Clinical Faculty of the University of Erlangen-Nürnberg (AZ: 04.03.10.1) as well as by the Deutsche Forschungsgemeinschaft (DFG), Sonderforschungsbereich 603, Teilprojekt C10.