Atlas-Assisted Tomography: Registration of a Deformable Atlas to Compensate for Limited-Angle Cone-Beam Trajectory

Abstract

We present a method to improve the quality of cone-beam tomographic images computed from an intra-operative C-arm scan by adding information from an anatomical atlas. Limited range of C-arm view angles leads to reconstruction artifacts and poor anatomical detail. We propose to complete the missing views with simulated projections of a statistical anatomical model, which is deformably registered to match the data in the C-arm images. This paper presents the methods used to create the atlas and to register it with x-ray images. We compare the results of seven leave-one-out simulated hybrid reconstruction tests on a population of 13 subjects, with "ground-truth" CT, a classical short-scan, and a partial-scan reconstruction.

Deformable 2D-3D Registration of the Pelvis with a Limited Field of View, Using Shape Statistics

“…In a previous publication [1], we showed that a deformably registered anatomical model can be combined with a limited-trajectory set of x-ray images to create a CT-like "hybrid reconstruction". The current paper focuses on assessing the accuracy of our deformable registration.…”

Statistical Atlases of Bone Anatomy: Construction, Iterative Improvement and Validation

“…Advantages of our approach include: 1) very minimal initial segmentation is required (once to create an initial mesh) and may be done manually or semi-automatically; 2) point correspondences are established automatically through intensity-based registration, avoiding landmark selection; 3) atlas bias/uncertainty is minimized through iterative refinement of an initial atlas; 4) the method requires little or no explicit prior anatomical information, although such information may be added in a separate annotation phase; and 5) the atlases produced are useful as prior information for assisting 3D/3D and 2D/3D (e.g., [5]) registration, as well as assisting in tomographic reconstruction from incomplete data [5].…”

“…Statistical characterization using principal component analysis (PCA) and point distribution models is presented in [1]. Following Cootes et al, a number of authors (e.g., [2], [3], [4], [5]) have applied similar methods to construct statistical atlases of bony anatomy from CT scans of multiple individuals. The basic method used is to identify landmark points, establish point-based correspondences between subjects, and then perform statistical analysis to study shape variations.…”

“…The basic method used is to identify landmark points, establish point-based correspondences between subjects, and then perform statistical analysis to study shape variations. Typically, subject anatomical shapes are represented as surface meshes [4] or volumetric tetrahedral meshes [2], [5], [3]. Often, a template mesh representing an anatomical structure is created, deformably registered to each subject, and mesh vertex points are used as the corresponding landmark points.…”

“…Wu et al created a statistical model using surface triangular meshes and a non-rigid point matching algorithm to register different shapes. Our atlas approach, originated by Yao et al [2] and subsequently also adopted by others [5] [3], represents the shape using tetrahedral meshes and registers the datasets using a grayscale deformable registration method. Bone density information is incorporated into the model using polynomials.…”