Multimodal Non-Rigid Registration for Image-Guided Head and Neck Surgery

PAGE Image guidance is a useful and enabling tool in the neurosurgical suite. The use of image guidance during brain tumor resection surgery provides the surgeon with detailed real-time anatomical information which has been shown to improve patient outcomes. Due to brain shift and surgical progression, the intra-operative anatomy may differ sig­ nificantly from the pre-operatively acquired images. Therefore, intra-operative images are acquired, but must be “registered” or fused, to the pre-operatively acquired im­ ages. We previously investigated the performance and efficiency of our state-of-the-art bio-mechanical non-rigid registration algorithm to register intra-operative MRI to pre­ operative MRI images in real time in a high-performance computing environment, and found th a t it improves registration accuracy between 3 and 8 times. In this study, we investigate the use of intra-operative Computed Tomography (iCT) as an intra-operative modality for non-rigid registration to pre-operatively acquired CT. The goal of this study is to evaluate the accuracy of our non-rigid registration algorithm using iCT images. Successful application of this technique would yield an intra-operative pseudo-MRI at a fraction of the cost. We implement two additional similarity metrics and develop and employ a synthetic benchmark which we use to both select a similarity metric and automatically evaluate the performance of the algorithm. To this end, and in participation with our clinical partners, we create and use a database of six patient cases. We find th a t pre-operatively acquired CT has sufficient definition to be used in our non-rigid registration algorithm, showing an average improvement of 4.14 times over rigid registration alone using synthetically-generated intra-operative images. We find, however, that our iCT are not suitable as an intra-operative modality currently, showing an average improvement of only 19% over rigid registration alone. We conclude that either more sensitive similarity metrics or an improved iCT is required for successful application of our non-rigid registration algorithm.