high performance computing on graphics processing units: hgpu.org

Localization of cortical regions of interests (ROIs) in the human brain via analysis of Diffusion Tensor Imaging (DTI) data plays a pivotal role in basic and clinical neuroscience. In recent studies, 358 common cortical landmarks in the human brain, termed as Dense Individualized and Common Connectivity-based Cortical Landmarks (DICCCOLs), have been identified. Each of these DICCCOL sites has been observed to possess fiber connection patterns that are consistent across individuals and populations and can be regarded as predictive of brain function. However, the regularity and variability of the cortical surface fold patterns at these DICCCOL sites have, thus far, not been investigated. This paper presents a novel approach, based on intrinsic surface geometry, for quantitative analysis of the regularity and variability of the cortical surface folding patterns with respect to the structural neural connectivity of the human brain. In particular, the Geodesic Field Estimate (GFE) is used to infer the relationship between the structural and connectional DTI features and the complex surface geometry of the human brain. A parallel algorithm, well suited for implementation on Graphics Processing Units (GPUs), is also proposed for efficient computation of the shortest geodesic paths between all cortical surface point pairs. Based on experimental results, a mathematical model for the morphological variability and regularity of the cortical folding patterns in the vicinity of the DICCCOL sites is proposed. It is envisioned that this model could be potentially applied in several human brain image registration and brain mapping applications.