Toward improved tumor targeting for image guided neurosurgery with intra-operative parametric search using distributed and grid computing

We describe a high-performance distributed software environment for real-time nonrigid registration during Image- Guided Neurosurgery (IGNS). The implementation allows to perform volumetric non-rigid registration of MRI data within two minutes of computation time, and can enable large-scale parametric studies of the registration algorithm. We explore some of the parameters of the non-rigid registration, and evaluate their impact on registration accuracy using ground truth. The results of the evaluation motivate running the registration process within the Grid environment, particularly when searching for optimal parameters of intraoperative registration. Based on the results of our evaluation, distributed parametric searching of optimal registration settings can significantly improve the registration accuracy in some regions of the brain, as compared to the accuracy achieved using the default parameters.