Segmentation of neuroblastoma in CT images using deformable contours, image reconstruction, and convex hulls

Segmentation of the tumor in neuroblastoma is complicated by the fact that the mass is almost always heterogeneous in nature. In this, paper, we propose to use deformable models to determine the tumor boundary on a slice-by-slice basis. The initial contour for the model is obtained from the convex hull of the thresholded result obtained by the application of opening by reconstruction to the original computed tomography data. For calcified tumors, the proposed method worked well; however, for diffuse tumors, the method resulted in under-estimated tumor regions.