A hybrid two-dimensional/three-dimensional atlas registration approach for medical images

A statistical shape atlas quantitatively represents the variations in shape of a set of training examples. This work presents a novel hybrid 3D shape atlas that consists of a global and local model, using a set of connected inscribed spheres that is contained within the bounding surface of the shapes. Unlike a statistical shape atlas constructed from current point distribution models, this hybrid atlas can be computed without specifying the correspondence of landmarks among the training examples and also does not need a global scaling transformation to align the training examples.; This work describes how to build a hybrid atlas and how to deformably register a new example shape to the atlas, either by using a set of 3D surface points or by using a few 2D X-ray images of the new example. This work also describes hardware-assisted 2D/3D intensity-based registration techniques that significantly decrease the computation time of the X-ray registration process. Both 3D and 2D atlas-based registration were experimentally tested, with registration accuracies that are suitable for use in computer-assisted orthopedic surgery.