Towards Exascale Parallel Mesh Generation and Real-Time Medical Image Computing

In this talk I will present a top down approach for guaranteed quality massively parallel mesh generation and its broader impact to real-time medical image computing. First, I will present a telescopic algorithmic approach to explore a billion-way concurrency by using a hierarchy of abstractions that leverage the memory and network hierarchy of current and emerging multi-layered parallel architectures. Then I will present our work on a parallel runtime software system for implementing effectively mesh generation codes with a billion-way concurrency. In the second part of this talk I will present a real-time, non-rigid registration method for aligning pre-operative MRI with intra-operative MRI to compensate for brain deformation during tumor resection. The new method leverages our work on parallel mesh generation and it reduces the alignment error up to seven and five times compared to a rigid and ITK’s FE-based non-rigid registration methods, respectively. On average, the alignment error of the new method is reduced by 9.23mm and 5.63mm compared to the alignment error from rigid and FE-based methods implemented in ITK. Finally, I will present preliminary data from the application of both technologies on Deep Brain Stimulation for Parkinson’s disease.