Referencing and Registration of Three – Dimensional Images
Two – dimensional (2D) cephalometric representations of the three – dimensional (3D) craniofacial structures cannot answer the many questions regarding treatment response mechanisms and localization of growth. 1,2 However, 3D assessments pose a significant challenge in the choice of superimposition of landmarks and/ or structures. Superimposition of serial radiographs onto a stable reference structure is necessary to evaluate growth or treatment change. 3D imaging from cone beam computed tomography (CBCT), axial CT, magnetic resonance imaging, or surface laser scans offers improved diagnostic information and, perhaps more importantly, also provides a better way to evaluate the changes created by treatment and the skeletal and soft tissue adaptive responses to treatment that occur.3 – 10 In 2D cephalometrics, the cranial base often is used for this purpose because it shows minimal changes after neural growth has been completed.
Although landmark location in 2D is hampered by the identification of hard and soft tissues on radiographs due to the superimposition of multiple structures, locating 3D landmarks on complex curving structures is significantly more difficult. As Bookstein noted,11,12 there are no suitable operational definitions for craniofacial landmarks in the three planes of space (coronal, sagittal, and axial). In the context of facial changes, superimposition should not rely on landmark identification or on best-fit techniques on structures that change over time with growth and treatment. Maxillomandibular changes need to be assessed relative to stable structures that have not been altered with growth or treatment.
To better understand soft and hard tissue facial forms and changes with growth and treatment, as well as to create normative databases and predict changes, registration on the whole surface of the cranial base is the best method. This chapter demonstrates the use of a fully automated voxel – wise rigid registration at the cranial base and the application of 3D superimposition methods to evaluate anatomic structures displaced by growth, surgery, or other treatment.
CBCT equipment specialized for maxillofacial imaging now offers a relatively low- dose and convenient way to follow changes in facial morphology in three dimensions. The use of 3D images for treatment planning and follow- up raises concerns regarding the radiation dose. CBCT scans of the cases presented were acquired with the NewTom 3G (AFP Imaging, Elmsford, NY, USA), with a 36.3 μ Sv acquisition dose for a maxillary and mandibular scan. This is a major reduction from 314 μ Sv for conventional axial CT. There is a decrease in the signal – to – noise ratio with CBCT, but there is complete visualization of the facial structures with spatial resolution of 0.36 mm in isotropic voxels. The imaging protocol utilized a 12 inch field of view to include the entire facial anatomy.
The analysis of serial CBCT images to evaluate changes over time is undertaken in a sequence of four steps, described below.
Construction of virtual 3 D surface models
Currently available commercial software does not allow the construction of virtual 3D surface models, but rather displays a 3D rendering that is a projection of the 3D structure of the face for visualization purposes. The implementation of registration tools using best-fit between the 3D renderings does not allow the quantification of local changes with treatment, but would allow an overall visualization. Longitudinal quantitative assessment of growth and surgical correction requires the construction of 3D surface models. The image analysis tools for this purpose are modifications of open source, freely available software from the National Institutes of Health (Bethesda, MD, USA).
Segmentation, that is, outlining the shape of anatomic structures visible in the cross- sections of a volumetric dataset from CBCT images, is performed with the ITK- SNAP tool,13 which is a freely available open- source program. ITK- SNAP is a tool for viewing 3D images and delineating and extracting anatomic structures, and it allows simultaneous visualization, navigation, and segmentation of all three planes (axial, sagittal, and coronal) with a linked cursor system that allows the tracking of a single voxel.
Segmentation can be completed using two different modes: semi – automatic segmentation and manual segmentation. In the semi – automatic mode, segmentation employs an algorithm that allows a deformable bubble to grow to define borders between neighboring anatomic structures. Many standard automatic segmentation methods fail when applied to the complex anatomy of patients with facial deformity. The methods described by Gerig and co- workers address these technical difficulties, and have been adapted by Cevidanes and co- workers to construct 3D craniofacial models.14,15
The 3D virtual models are usually built from a set of around 300 axial cross- sectional slices for each image, with the voxels reformatted for an isotropic of 0.5 × 0.5 × 0.5mm. This resolution is used because although higher spatial resolution with smaller slice thickness is possible, it increases the image file size and requires greater computational power and user interaction time without significantly improving the quality of assessment of changes between time points. After segmentation with the ITK-SNAP tool, these files are converted from volumetric data into surface meshes for the 3D shape analysis procedures.
Registration of longitudinal 3D CBCT images and surface models
This is a core technology to assess growth and treatment outcomes that can involve nonrigid or rigid registration procedures. Nonrigid procedures can be used to assess across patient differences in size and morphology. The two obstacles to the widespread clinical use of nonrigid (elastic and deformable) registration are computational cost and quantification difficulties as the 3D models are deformed. Nonrigid registration would be required to create a composite of several different jaw shapes to guide the construction of template or standard normal 3D surface models. However, to evaluate longitudinal changes, a rigid registration of CBCT images is acceptable, and establishes a common coordinate system in which longitudinal changes can be assessed.
To use only stable structures and avoid observer- dependent techniques based on anatomic landmark identification, we mask anatomic structures displaced with growth and/or treatment and then perform a fully automated voxel- wise rigid registration at the cranial base, using rigid registration and Imagine software.16 Rotation and translation parameters are calculated and then used to register 3D models from before and after treatment on the cranial base (Figures 6.1 and 6.2).
For the registration of CBCT scans of subjects in whom cranial base growth is complete, registration of virtual 3D surface models is done utilizing the whole surface of the cranial base (Figure 6.3). To evaluate within- subject changes with growth and treatment, the endocranial surfaces of the anterior cranial fossae and the ethmoid bone are used in the registration procedure, since the growth of these structures is completed in early infancy. In this way, the anterior cranial base of the CBCT images is used as the reference for superimposing different time points. Maxillomandibular changes are described not as absolute displacement, but as displacement relative to the cranial base (Figures 6.3 and 6.4).
Superimposition of 3D surface models
After registration, the overlay of the 3D models can be assessed using a number of different publicly available tools, such as Slicer3, Vol2Surf, FltkSOV3Dtool, Valmet, and MeshValmet developed by the National Institutes of Health roadmap National Alliance of Medical Computing, of which University of North Carolina (UNC) is part.
The next step in the analysis involves overlaying the 3D model surfaces that are registered in the same coordinate system with another tool, CMF software (Maurice M üller Institute, Bern, Switzerland).17 This tool allows different degrees of transparency to visually assess the boundaries of the maxillomandibular structures between superimposed models at two different time points. This clearly identifies the location, magnitude, and direction of mandibular displacements, and also allows a quantification of the vertical, transverse, and anteroposterior bone displacements and remodeling that accompany growth and response to treatment.
Visualization and quantitative assessment of changes with growth and t reatment
Precise quantitative measurement is required to assess the changes resulting from growth and treatment. For example, in the evaluation of surgical treatment, we need to identify the placement of bones in the desired position, the sites of surgical cuts, fixation of screws, and/or plates relative to risk structures, and the location and amount of post-treatment bone remodeling. Landmark- based measurements reflect errors related to landmark identification. The use of semi-landmarks has been proposed, that is, landmarks plus vectors and tangent planes that define their location, but information from the whole curves and surfaces must also be included.18,19