Objective evaluation of accuracy of intra-operative neuroimage registration

Abstract

Pre-operative brain images that are registered onto relevant intra-operative images can enhance navigation during image-guided neurosurgery. One of the crucial steps in the process of image registration is assessment of its accuracy. The accuracy of an image registration procedure was evaluated in one of our previous studies, for five cases of neurosurgery, using a manual segmentation-based method that is subjective and prone to human errors. The aim of this study is to develop an evaluation method that is objective and automatic. An edge-based Hausdorff Distance (HD) metric based on Canny edges was developed for evaluation. Subsequently, the accuracy of non-rigid registration (NRR) results was evaluated using intra-operative images as ground truth and compared with those from the previous study. The obtained results compared well despite the differences in the methods employed. The edge-based HD metric provides an objective measure for image registration accuracy evaluation.