A rapid 3D reconstruction of bones and other structures during an operation is an important issue. However, most of existing technologies are not feasible to be implemented in an intraoperative environment. Normally, a 3D reconstruction has to be done by a CT or an MRI pre operation or post operation. Due to some physical constraints, it is not feasible to utilise such machine intraoperatively. A special type of MRI has been developed to overcome the problem. However, all normal surgical tools and instruments cannot be employed. This chapter discusses a possible method to use a small number, e.g. 5, of conventional 2D X-ray images to reconstruct 3D bone and other structures intraoperatively. A statistical shape model is used to fit a set of optimal landmarks vertices, which are automatically created from the 2D images, to reconstruct a full surface. The reconstructed surfaces can then be visualised and manipulated by surgeons or used by surgical robotic systems.
Top2. Background
In an intraoperative environment where the scanning geometry of a CT or MRI is not suitable, a C-arm conventional X-ray system can be used to acquire a number of 2D images to reconstruct a full 3D volumetric description, in terms of voxels, of an object of interest, e.g. (Atesok, et al., 2007; Ritter, Orman, Schmidgunst, & Graumann, 2007; Zbijewski & Stayman, 2007). However, to reconstruct at a reasonable resolution, the number of 2D images required is very high, e.g. 40 to 180 images, and to extract the surface of an object from the reconstructed voxels is very computationally expensive. Moreover, due to the large number of 2D X-ray images required, the patient is inevitably subjected to high dose of radiation.
Another approach is to use statistical shape analysis and modelling, e.g. (Cootes, Taylor, Cooper, & Graham, 1995; Dryden & Mardia, 1998; Kumar T. Rajamani, et al., 2007; Guoyan Zheng, Dong, Rajamani, Zhang, & Styner, 2007), which has been an important tool in 3D model reconstruction from incomplete data. In this approach, only a small number of sparse landmark vertices on the surface of an object, e.g. a bone, are needed to be determined. Those sparse landmark vertices alone contain inadequate information for the complete 3D surface reconstruction of an object. Hence, a priori knowledge is required. A statistical model can be reconstructed from a set of training surfaces representing reasonable variations of the surfaces of an object of interest. In intraoperative applications, the statistical model is then used as prior knowledge in the reconstruction process to fit to the patient anatomy using intraoperatively acquired sparse landmark vertices. Thus, in conclusion, the aim of statistical shape model fitting is to extrapolate from an extremely sparse and incomplete set of 3D landmark vertices to a complete and reasonably accurate 3D anatomical surface. The fitting process aligns and deforms the statistical shape model to fit the sparse landmark vertices. Therefore the model-based approach is widely accepted due to their ability to effectively represent objects.