Non-Rigid Registration for Image-Guided Neurosurgery on the TeraGrid: A Case Study

We present a feasibility study of using TeraGrid resources for computing non-rigid registration (NRR) of brain MRI. The end-to-end NRR procedure we consider has previously undergone clinical validation at Brigham and Women’s Hospital, Boston. We consider the use of TeraGrid resources in two scenarios. First, we evaluate the feasibility of using a grid-enabled implementation to provide timely execution of the most time-consuming components of the registration. Second, we describe a workflow implementation to enable speculative computation of registration to improve confidence in the result and assist in retrospective evaluation of the method. Our initial results indicate that TeraGrid provides practical and cost-effective means to support IGNS image processing. At the same time, performance limitations of the intra-operative speculative NRR execution using conventional means of workflow scheduling prevent its intra-operative applications. The results of this paper motivate future work in highthroughput scheduling and execution models on TeraGrid, which can be leveraged by the workflow implementation we present.