Performance Evaluation of the Modified Iterative Closest Point Methods for Intra-operative Ultrasound and pre-operative MR Image Registration of Brain

Intra-operative ultrasound imaging as a non-ionized and being real time has been found very applicable as an intra-operative update of patient data in image guided neurosurgery system. The main point is the accurate registration of intra-operative with pre-operative images. Due to speckle noise in ultrasound images, scale differentiation between MR and ultrasound images and their different resolution, an accurate registration of ultrasound images with pre-operative MR images is a challenging problem. In this paper the effect of different steps of the Iterative Closest Point is considered and, then, the best modified version of ICP is introduced for this type of data. To perform this study, a Poly Vinyl Alcohol-Cryogel brain phantom is used which allows simulating brain deformation. The performance of the best version of ICP is compared to a well-known point based algorithm, Coherent Point Drift in terms of accuracy and speed. The results proved CPD algorithm was more robust than ICP algorithms in the presence of noise, although with a more computational cost. Changing different steps in conventional ICP has led to improve the performance of the ICP. As the results of our phantom study confirm the best version of ICP has not only achieved an accuracy close to CPD method, but also in a much faster approach. According to a trade off between the speed and accuracy of nine implemented versions of ICP algorithms, using some modified version of ICP is preferred to CPD method.