Connectome-GPU

The neurobiological data sets in this project are volumes of extremely high resolution that are created by electron microscopy and confocal microscopy, resulting in multiple to hundreds of terabytes per volume. The ultimate goal is to determine the detailed connections of the contained brain circuits, which is a fundamental unsolved problem in neuroscience.

Understanding this neural circuitry will enable brain scientists to confirm or refute existing models, develop new ones, and come closer to an understanding of how the brain works. This is the largest-scale visual computing problem of which we are aware, approximately five orders of magnitude beyond current state of the art.

The Harvard Center for Brain Science (CBS) has been able to achieve very promising early results, but a major obstacle going forward is handling the enormous size of the acquired data sets. The Initiative in Innovative Computing (IIC) at Harvard is tasked with developing software and hardware solutions for this problem, and is cooperating with VRVis and Microsoft to develop a scalable client-server solution employing a GPU (graphics processing unit) cluster for both visualization and computation.

The system builds on scalable multi-resolution approaches throughout and exploits the parallel processing power of GPUs, which are an extremely powerful and affordable platform for high-quality volume rendering, especially using raycasting, and general purpose scientific computations. The developed visual computing system will allow scalable volume rendering and semi-automatic segmentation of volumes in the hundreds of terabyte range with interactive feedback. It will also be applicable to a variety of other large data problems, such as visual computing for astrophysics, other kinds of biomedical data, and large time-dependent volume data.