Massively parallelized brain tractography using compute clusters, supercomputers, and graphics processing units

Abstract

Tractography based on diffusion-weighted MRI (dMRI) is the predominant in vivo method for mapping the brain’s white matter. However, it is also one of the most computationally demanding steps in neuroimaging data analysis — requiring the generation and filtering of millions of streamlines per subject. Over the past decade, high-performance computing (HPC) and graphics processing units (GPUs) have reshaped the landscape of tractography. What once required hours or days per subject can now be performed in minutes or even seconds. This facilitates faster processing of large-scale studies and high-resolution datasets, supports real-time clinical applications, and enables more computationally demanding algorithms. This review surveys the recent wave of applications of HPC and GPUs to massively parallelize tractography. We discuss how these advances have accelerated tractography pipelines, identify common strategies for parallelization, and highlight opportunities where further parallelization could improve efficiency and accuracy.