Many segmentation techniques exist in the literature. Well known is global thresholding with automated threshold selection based on the image histogram, e.g. Otsu's method. Other commonly used methods are region growing, watershed segmentation, active contours, etc. In the setting of CT, these techniques base themselves exclusively on the tomographic reconstruction or the tomogram. In practice, however, these tomograms will not be completely accurate because of reconstruction error and artefacts (e.g. due to noisy or insufficient data) and the useof classic segmentation techniques will thus also lead to inaccurate segmentations. Therefore, we develop objective segmentation methods that not only use the tomogram, but also the available projection data.
One such technique is the Projection Distance Minimization method (PDM). For a given threshold (either global or local), partitions are defined. Then, for each partition the grey level is estimated in a way such that the projection distance will be minimal. A forward projection of this segmentation will be created and its euclidean distance to the real projection data will be measured. The optimal segmentation is the one such that this distance is minimal. Experiments have shown that this method indeed creates superior segmentations when compared to classic techniques such as Otsu's. The difference is mostly noticeable when the amount of projection angles is small. This means that high accuracy segmentations can be made while limiting the dose used in the tomographic process. This is very important in biomedical studies on rats or mice.
Another technique is the Segmentation Consistency Maximization method (SCM). Most segmentation techniques assume that the number of different partitions in the image domain is limited. This is not always the case and often one is interested in finding the object with the highest, homogeneous density (e.g. a metallic implant) amidsts objects of a varying densities (e.g. bones and tissue). The SCM-algorithm finds a global threshold to distinguish between the metallic object and the background by minimizing the A-incosistency of the residual sinogram. The latter is the projection data minus the forward projection of the segmentated dense object. The concept of A-inconsistency means the amount of inconsistency of a sinogram given the fact that a number of pixels (those belonging to the dense object) may not be used in a reconstruction. This value is measured by creating an iterative reconstruction of the residual sinogram and comparing its forward projection to the residual sinogram.