Abstract
This report presents a new method for non-rigid registration between an organ surface model and given 3D X-ray CT images. The method represents an organ surface by using a statistical point distribution model, and registers the model by estimating the marginal posterior probability distribution of the location of each of the points. The posterior distributions show not only the location of the registered surface but also the pointwise confidence of the estimated locations. Each point in the model has three probability distributions: the prior distribution of the location of each point, the conditional probability distribution of the local appearance around each point, and the probability distributions of the relative positions between two neighboring points. When a new image is given, we estimate the posterior distributions by means of a non-parametric belief propagation. Experimental results showed that the method works successfully with clinical images.
