Model-Based Organ Segmentation in CT Scans

Figure Number Page 1.1 CT scans (axial view) and liver segmentation. The first column shows the CT scans and the second column shows the corresponding ground truth liver segmentation.. 2.2 A set of training CT volumes. The colors represent the intensity values in the CT volumes. The shape of an organ is represented by a triangular mesh model. A triangular mesh that contains three vertex points, P 1 , P 2 and P 3 , is drawn in each organ. For each vertex point, an arrow is drawn to represent its normal vector and a 3D box represents its intensity profile. We use different sizes and colors to indicate the variations in sizes and intensities of organs.. . 7 2.3 3D point correspondence. Points in the same color indicate they are corre-2.4 Learning a statistical shape model. A triangle represents a training shape and the mean shape of all the training shapes is drawn with a diamond shape. Here we assume that we learn a linear model (e. 2.5 Learning a boundary intensity model. A simple histogram model indicates the probability of a point belonging to the boundary given its intensity profile. 9 iv 2.7 Boundary refinement.(a) For a vertex point, we draw its normal vector as a red arrow and the search range as a violet box. The learned boundary intensity model is used to estimate the refinement (e.g., estimating a displacement vector for each vertex point in the search range or the violet box of the current vertex point).(b) After the displacement for each vertex point is estimated, we can get an updated shape. (c) The updated shape is shown in the shape space as a blue square. (d) The closest shape to the updated shape (the blue square) on the statistical shape model (the dotted line) is drawn as the green circle. (e) The closest shape is put in the CT volume. (f) The initial shape is replaced with the (closest) shape and the process (steps a-f) is repeated.