Rigid registration of 3D ultrasound data of brain tumours

Abstract

Intraoperative brain deformation is one of the most contributing factors to the inaccuracy of image-guided surgery systems. One option for correcting this deformation is to acquire 3D ultrasound images during surgery and using these to update the information provided by the preoperatively acquired MR data. To compare ultrasound volumes acquired at different stages of surgery, non-rigid registration techniques are necessary. As a first step towards this approach, the accuracy and robustness of entropy-based rigid registration methods for ultrasound volumes have been examined. From the robustness measurements on patient data, it could be concluded that the entropy-based methods mutual information and normalized mutual information (NMI) with a number of histogram bins of 64 or 128 using partial volume interpolation perform robustly for rigid registration of ultrasound volumes of brain tumours. When no brain deformations occur between the two ultrasound acquisitions, accuracies of around 0.5 mm can be achieved.