22 January/February 2014 Published by the IEEE Computer Society 0272-1716/14/$31.00 © 2014 IEEE Feature Article GPU-Accelerated Interactive Visualization and Planning of Neurosurgical Interventions Mario Rincón-Nigro, Nikhil V. Navkar, Nikolaos V. Tsekos, and Zhigang Deng ■ University of Houston U sing preoperative imaging to plan in- terventional procedures has led to new, effective interventional paradigms. Neu- rosurgical procedures that beneit from such plan- ning include deep brain stimulation, biopsies, and shunting and insertion of external ventricular drains. Planning often entails processing multi- contrast and multimodal imaging (magnetic reso- nance imaging and computed tomography) to map anatomical and pathophysiologic features. A common practice is manual selection of different entrance positions on the patient’s scalp and assessment of their suitabil- ity, to determine an appropriate insertion path. To avoid errors due to discrep- ancies between preoperative and intraoperative images, the trend is to move as much of the plan- ning as possible into the operat- ing room. Toward that end, we propose a semiautomated GPU- accelerated method to process, visualize, and plan interventions at interactive or nearly real-time speed. It has two main components: ■ It embeds the geometrical structures represent- ing critical-tissue areas pertinent to the proce- dure in spatial data structures. This speeds up computation of the geometric queries involved in estimating the risk for paths. ■ It implements computation on GPUs, which ex- ploits the problem’s parallel nature while effec- tively handling the involved irregular workload. Evaluations have demonstrated that our method is robust and interactive and can generate safer straight access paths. Figure 1 diagrams the method. To the best of our knowledge, our method is the irst to enable interactive estimation and visualiza- tion of risk in straight-access neurosurgical inter- ventions with mesh-based tissue representation. Owing to its high speed, the operator can inter- actively explore a much larger number of possible paths and target points. This ability enhances the effectiveness and eficiency of decision making in interventions. Applying Our Method The operator irst selects a target point in a predeined region—for example, in the tumor’s core or on its surface. He or she then interactively visualizes the resulting risk map to guide selection of an optimal insertion point. Figure 2a illustrates this process. After selecting a target point and an insertion point, the operator can view the risk map for the advancing needle’s current position (see Figure 2b). Such dynamic risk maps can signiicantly fa- cilitate decision making. When selecting a path, the operator usually further analyzes it by looking at imaging planes (magnetic-resonance-imaging slices) orthogonal to it (also called a bird’s-eye view). For stereotactic robot-assisted neurosurgi- cal interventions, such as in the NeuroArm and A proposed GPU-accelerated method enables interactive quantitative estimation of the risk associated with neurosurgical access paths. It exploits spatially accelerated data structures and eficient implementation of algorithms on GPUs. In evaluations, the method achieved interactive rates, even for high-resolution meshes.