On the performance of improved ICP algorithms for registration of intra-ultrasound with pre-MR images; a phantom study

Abstract

Ultrasound imaging as a simple and being real time has been found very applicable for intra-operative updates of pre-operative MRI data in image guided neurosurgery system. The main challenge here is the presence of speckle noise which influences the accuracy of registration of US-MR images, intra-operatively. In this paper the performance of two improved versions of the well known Iterative Closest Point (ICP) algorithms to deal with noise and outliers are considered and compared with conventional ICP method. To perform this study in a condition close to real clinical setting, a PVA-C brain phantom is made. As the results show improved versions of ICP are found more robust and precise than ICP algorithms in the presence of noise and outliers. Then the effect of various de-noising methods including diffusion filters on the accuracy of point-based registration is evaluated. The role of a proper diffusion filter for de-noising of US images has also improved the performance of the ICP algorithm and its variants about 35% and 20%, respectively.