GPU Technology Conference

2652's picture
Research Title: 
Computational Modeling of Human Head Electromagnetics for Source Localization of Milliscale Brain Dynamics using GPUs
Authors: 
Adnan Salman
Authors: 
Sergei Turovets
Authors: 
V. M. Volkov
Authors: 
A. Malony
Authors: 
Don M Tucker
Authors: 
P. Luu
Country: 
USA
Research Abstract: 

Advances in human brain science have been closely linked with new developments in neuroimaging technology. Brain activity takes place at millisecond temporal and millimeter spatial scales through the reentrant, bidirectional interactions of functional neural networks distributed throughout the cortex and interconnected by a complex network of white matter fibers. Our research goal has been to create an anatomically-constrained, spatiotemporally-optimized neuroimaging (ACSON) methodology to improve the source localization of dense-array EEG (dEEG). Anatomical constraints include high-resolution three-dimensional segmentation of an individual's head tissues, identification of head tissue conductivities, alignment of source generator dipoles with the individual's cortical surface, and interconnection of cortical regions through the white matter tracts.  Using these constraints, the ACSON constructs a full-physics computational model of an individual's head electromagnetics and uses is to map measured EEG scalp potentials to their cortical sources. The ACSON workflow (see diagram) poses several major computational challenges to be applied in practice.  High-performance parallel implementations of the electromagnetic solvers using GPUs has enabled the creation of one of the first full-resolution, FDM construction of a real human head for source localization based on electromagnetics simulation Computational Modeling of Human Head Electromagnetics for Source Localization of Milliscale Brain Dynamics using GPUs.