Simulation of Stomach Specimens Generation Based on Deformation of Preoperative CT Images

Abstract

This paper presents a novel method for generating virtually-unfolded views of the stomach by cutting and deforming CT images. Unfolded views are very useful for the diagnosis and treatment planning of stomach cancer, since they provide various information of the lumen which can only be obtained from a resected specimen. However, conventional methods cannot correctly reproduce luminal surfaces because elasticity for the shape model of the stomach is quite coarse defined. In this paper, we use Voigt elements for elasticity modeling, forces calculated from surface normals for directing the stomach to a flat shape, and the Newmark-β method for image deformation. We simulated deformation of a phantom dataset and compared the stability as well as computation time with the Euler method. Unfolded views from fifteen CT image datasets, corresponding virtual gastroscopic images, and resected specimens were used for comparison with conventional methods. Experimental results showed that our method can generate views faster in which concave regions of the stomach are better flattened and 99% of the luminal surface can be reproduced. Unfolded views from twelve datasets were presented for surgical planning. They were considered to have well reproduced lesions as well as fold patterns observed in virtual gastroscopic images and resected specimens.