LIVER IMAGE REGISTRATION BY FINITE ELEMENT MODEL FOR DERIVING TUMOUR AND VESSEL LOCATIONS

Abstract

This research proposes a marker-less surface alignment method in finite element model-based image registration. The objective is to develop a non-rigid image registration process which can be used to derive the intraoperative liver tumour and vessel locations to assist surgeons in removing liver tumour without injuring large blood vessels in minimally invasive surgery. Before the liver resection surgeries, surgeons investigate liver anatomy, tumour location and large blood vessel locations from the preoperative liver images to form surgical planning. During minimally invasive surgeries, surgeons may perform many kinds of surgical operations on the liver in order to remove the liver tumour. The operations usually cause severe deformations of the liver which make it difficult to identify the accurate locations of the tumour and the vessels. The removal of the tumours located in the posterior of a liver or close to large blood vessels may run the risk of injuring the tumour or the large vessels while resection. This research proposes an approach to derive the tumour and vessel locations to avoid the above-mentioned problem. In this research, preoperative biomechanical mesh models are constructed by CT scans. Intraoperative surface models are built by the contours of intraoperative images. Surface alignment strategy adopts the normal projection of the minimal distance vector and the optimized constraints to iteratively minimize the difference between preoperative and intraoperative models. Finite element model based on surface alignment strategy to deform preoperative mesh model, such that it aligns with intraoperative surface model to get intraoperative tumour and vessel locations. The proposed method is first validated by simulations of a surgical scenario and then validated by the deformation experiment on the ex vivo porcine liver. The root mean square errors of the liver surface marker and the internal marker for simulations are 0.35 ± 0.43 mm and 0.28 ± 0.20 mm, respectively. The fiducial registration errors for the liver surface and the internal for the ex vivo porcine liver experiment are 2.44 ± 1.02 mm and 5.48 ± 4.57 mm, respectively.