High-Performance Correlation and Mapping Engine for rapid generating brain connectivity networks from big fMRI data

Abstract

Brain connectivity networks help physicians better understand the neurological effects of certain diseases and make improved treatment options for patients. Seed-based Correlation Analysis (SCA) of Functional Magnetic Resonance Imaging (fMRI) data has been used to create the individual brain connectivity networks. However, an outstanding issue is the long processing time to generate full brain connectivity maps. With close to a million individual voxels in a typical fMRI dataset, the number of calculations involved in a voxel-by-voxel SCA becomes very high. With the emergence of the dynamic time-varying functional connectivity analysis, the population-based studies, and the studies relying on real-time neurological feedbacks, the need for rapid processing methods becomes even more critical. This work aims to develop a new method which produces high-resolution brain connectivity maps rapidly. The new method accelerates the correlation processing by using an architecture that includes clustered FPGAs and an efficient memory pipeline, which is termed High-Performance Correlation and Mapping Engine (HPCME). The method has been tested with datasets from the Human Connectome Project. The preliminary results show that HPCME with four FPGAs can improve the SCA processing speed by a factor of 27 or more over that of a PC workstation with a multicore CPU.