Internal Motion Estimation during Free-Breathing via External/Internal Correlation Model

Determination the locations of vessels and tumors are the key components in radiofrequency ablation (RFA). Traditionally, patients are required to hold their breath during puncture for precisely locating their target structures. However, it's difficult for patients to do so since it is hard to avoid the free breathing completely. In this paper, we propose a novel external-internal correlation model to predict the liver surface first, and then estimate the locations of the blood vessels and tumors inside the liver during free-breathing via non-rigid registration. 4D-CT scanner is used to obtain the external and internal motions simultaneously, here we represent the body surface with 6 surface landmarks, while reconstruct the 3D model of internal liver surface. After that, we apply the moving least squares (MLS) method on surface landmarks to predict the partial motion of liver surface. Considering that laparoscopic stereo stream has been successfully employed to register the given preoperative volume on liver surface, we adopt the pre-trained convolution neural network (CNN) on partial liver surface data to get the displacement of heterogeneous liver, which can be used to estimate the motion of internal structures. The experimental results show that our method can well predict the displacement of interesting internal liver structures.