Parallelization of COMMIT using CUDA : Acceleration with GPU of a large-scale problem for microstructure informed tractography

Abstract

This document presents an optimization of the COMMIT framework develop ed by the Dr. Alessandro Daducci and his collaborators. COMMIT framework has been used to filter tractograms which are useful to study the connections in the human brain. We parallelized with the CUDA language the algebraic op erations Ax and Aty in order to accelerate the optimization pro cedure necessary to filter a tractogram with COMMIT. The results of our parallel implementation of the operations were validated by comparing the results with the current version of COMMIT. This work shows exp eriments with real human brain data which demostrate that the parallel versions of the op erations Ax and Aty significantly reduced the computational time required to filter a tractogram. This thesis contribute with a faster version of the COMMIT framework which uses a NVIDIA GPU to accelerate the op erations Ax and Aty along with backward compatibility with the previous COMMIT scripts.