3D segmentation based on cylindrical B-spline active surface model

Reconstruction of the surface of left ventricle (LV) is one of the challenging problems to analyze left ventricle's shape and motion. In this paper, a cylindrical B-spline active surface (CBAS) model is put forward. The model can be used to segment the LA (long axis) and SA (short axis) tagged MR images to get the 3D surface of the LV. The model can be seen as a grid composed of B-snake contours. According to the relation of LA and SA image plane's position in the 3D space, one node of the grid synchronously finds the edge points in one SA image and one LA image, and the sample points between two nodes in the same direction calculate their image energies using only one SA or LA image. By using the cylindrical model, the deformation of the model on the SA planes or LA planes can be realized just by changing only one variable, which decreases the complexity of the model. In the paper, firstly, the fuzzy Hough transform is used to find the approximate position of the LV's contour. Secondly, the approximate contours are used to construct the initial surface of the CBAS model. Then, the model is utilized to segment the tagged MR image series and get the exact LV's surfaces. The experiment shows that the method can get exact 3D surfaces by being compared with manually outlined contours.