The Effect of Segmentation Method on the Performance of Point Based Registration of Intra-Ultrasound with Pre-MR Images

Intra-operative Ultrasound imaging as a real time imaging device is found very applicable for intraoperative updates of patient data in neurosurgery. One of the main challenges here is the accuracy of intra and preoperative image registration which is highly influenced by the level of speckle reduction and the accuracy of segmentation method. In this paper the effect of segmentation method on the accuracy of point-based registration of intraoperative US image with preoperative MR image using Iterative Closest Point (ICP) algorithm and Coherent Point Drift (CPD) are considered. To perform this study, a Poly Vinyl Alcohol-Cryogel brain phantom is made that allows simulating the brain deformation. As the results showed CPD algorithm is found more robust than ICP in the presence of outliers. In addition, the role of Chan-Vese as a nonparametric active contour based segmentation of US images has shown a reasonable improvement in the performance of the conventional ICP and CPD in order of 45% and 30%, respectively.