Uncertainty in Image-to-Physical Registration for Soft-Tissue Image Guided Surgery

Abstract

Image-guided surgical methods have been investigated as techniques to increase the localization accuracy of hepatic cancer treatments such as resection and ablation. Mathematical registration methods are currently used to align intraoperative physical space with preoperative image space, providing the physician with more information in the surgical setting. These methods rely upon accurate digitization of the intraoperative organ surface. Across the data collection of a series of clinical cases, we observed a high variability in the pattern and density of acquired surfaces. The goal of this work was to characterize the extent to which variation of input data affects the output of image-to-physical registration methods. In order to do so, a data set consisting of multiple realistic surface acquisitions of the same intraoperative organ was realized by virtually projecting the collection pattern of 14 clinically acquired hepatic surfaces onto an anthropomorphic liver phantom. With this simulated set of data, we observed that varying surface collection has an effect on the accuracy and repeatability of registration methods. In addition, a strategy for normalizing, or resampling, collected surface data was developed and applied to the simulated data sets. Results of this work suggest (1) the technique of surface acquisition has downstream effects on registration error and (2) a surface resampling strategy may be used to normalize data acquisition across cases, and users, to further increase the accuracy of current clinical methods.