3D Model Superimposition for CBCT Alignment

3D models were constructed and superimposed using voxel based superimposition in Maxilim software installed on a windows XP-based workstation (Intel® core™ 2 Duo; 2.9 GHz, 3.25GB, ATI Radeon™ 3450 HD graphics card). The construction of the 3D models was performed by selecting the range of Hounsfield unit (HU) representing the bony tissues on the DICOM images. This was achieved by selecting a lower threshold value between 250–350 HU. Values above this threshold were automatically selected. The superimposition procedure is an automated procedure that compares the grey values in the two DICOM images voxel by voxel. The user is first required to select the volume of interest (registration area), then to roughly align the 3D models. Consequently the software computes the translation and rotation needed to geometrically align the two DICOM images, and subsequently the constructed 3D models, based on the maximization of mutual information. For each pair of CBCT scans the 3D model construction and superimposition procedure was repeated five times with a time interval of three weeks.
The scans were registered twice on the anterior cranial base and twice on the left zygomatic arch (zygomatic bone + zygomatic process of the temporal bone) by the same operator (RN) (Fig. 1). To test the inter-observer reliability, the scans were superimposed for a fifth time by a second observer (HB) registered on the anterior cranial base.