Computation cloud to enable high throughput neuroimaging

Abstract

Neuroimaging studies require significant computational power in order to perform non-linear registrations, 3D volumetric segmentations, and statistical analysis across a large group of subjects. In addition to the need for this large computational infrastructure, the large number of open-source programs being used to process data has increased in recent years making it standard for several packages, each one frequently and independently updated, to be used in a single analysis. Due to these needs, the focus of computational infrastructure in neuroimaging is transitioning from user-owned hardware, to virtualized, shared, and scalable “cloud”-based hardware. We have implemented such a “private cloud” for neuroimaging and deployed it to users of the Beckman Institute Bioimaging Center. This thesis aims to demonstrate the scientific advantage for neuroimaging from such a system, to serve as a guide for users and administrators, and to provide implementation details to other groups who may wish to build a similar cloud of their own. In the final chapter, we present a sample application—a novel, open source method for the detection and quantification of multiple sclerosis lesions on MRI images. Like many neuroimaging applications, this method takes a great deal of time to process all subjects, highlighting the practicality of the flexible, shared infrastructure of the private cloud for bioimaging research.