ABSTRACT
Image registration is used to determine the spatial relationship between two image data sets. Images acquired by the same (i.e., intramodal) or different (i.e., intermodal) imaging modalities can be used for (i) the sole purpose of determining the spatial transformation between two imaging data sets usually differing temporally or (ii) combining information from two different image data sets differing temporally or by modality. In radiation therapy, image registration is used in treatment planning, treatment assessment, and patient-position verification. In treatment planning, image registration is used to align multiple image data sets (both intramodal and intermodal) to assist the physician in contouring anatomical structures and planning target volumes. In treatment assessment, intermodal imaging [e.g., computed tomography (CT) and cone-beam computed tomography (CBCT)] can be used to assess dose delivery in terms of intended dose (i.e., treatment planning) and delivered dose (i.e., treatment delivery). In patient-position verification, image registration is used to verify patient positioning before or during treatment. Rigid registration involves a single global displacement vector and/or rotational transformation applied to an entire image data set. It is used in treatment planning, treatment assessment, and patient-position verification. Deformable registration involves local transformations applied to an image data set in cases where
the patient may have undergone physical changes that render a rigid registration anatomically inadequate. It is used in treatment planning and treatment assessment. Whether rigid or deformable, image registration is the process of combining different imaging data sets onto a common coordinate system where there is a mapping between the voxels (i.e., pixels) of the two data sets. This process involves a fixed or reference data set (i.e., image) and a moving or target data set. The target is aligned to the reference data set. (It should be noted that in literature that one sometimes finds the terminology of source and target sets used, with the source indicating the moving image and the target indicating the fixed or reference image. We will not be using this terminology here.) Rigid registration involves a single displacement vector, consisting of three orthogonal translations, and three angular rotations, which are globally applied to the target image so that the same transformation is applied to each voxel in the target image. Deformable registration involves aligning the target image to the reference image, whereby each voxel is assigned a displacement vector; this displacement vector can be different for the neighboring voxels so that the shape and volume can be altered.
