A GPU-Based Parallel Computing Platform for High-Speed Brain Anatomical Connectivity Analysis

The functional areas in cerebral cortex communicate with each other by releasing and transportation of the neurotransmitter through the neural axonal fibers. Based on the established anatomical and functional connectivity, the interaction of coordination, collaboration and inhibition of the involved functional areas play an important role in execution of specific cognitive functions. Thus, by investigating the correlation between measured connectivity and behaviors, we could gain much insight into how brain works and communicates internally. Recent advances of non-invasive diffusion MRI technique provide a great opportunity to explore the human brain anatomical connectivity. By probing three-dimensional displacement distribution of diffusive water molecules within fibrous brain structure, diffusion MRI has shown its capability to reflect the microstructural components and neural direction by measuring the diffusion probability density function and orientation distribution function. Linked with neural tracking methods, neural tractography in living human brain opens a window to non-invasively investigate the white matter pathways. The aim of this study is to develop an integrated platform for brain anatomical connectivity analysis with high angular resolution diffusion imaging (HARDI) based on the graphic processing units (GPUs) and compute unified device architecture (CUDA). With the ability of HARDI for resolving heterogeneity of white matter fibers within a voxel, the massively parallel processing power and excellent floating point performance of the GPU, and a comprehensive development environment for C and C++ by CUDA, this platform will provide three modules, including the high-speed processing procedure for the image reconstruction of diffusion MRI, the extraction and analysis of tractography, and real-time and interactive interface for three-dimensional visualization. In summary, fulfilling this project would shed a light on demonstrating a high-speed medical image processing platform for brain anatomical connectivity analysis, presenting following advantageous features: (1) saving the time between data acquisition of diffusion MRI and extraction of fiber trajectories and facilitating the clinical assisted diagnoses and real-time guideline for neurosurgery; (2) providing the high-speed and accurate delineation of brain structural connectivity by GPU-based tractography algorithm and HARDI data; (3) estimating an integrated neuroscience tool with the ability of GPUs parallel computing and the functions of HARDI reconstruction, neural fiber tracking and visualization for the academic studies on neuroscience and brain research and the future industrial collaboration.

Project ID:PB10106-0311
External Project ID:NSC101-2218-E182-005