Knowledge-based automated reconstruction of human brain white matter tracts using a path-finding approach with dynamic programming

Abstract

It has been shown that the anatomy of major white matter tracts can be delineated using diffusion tensor imaging (DTI) data. Tract reconstruction, however, often suffers from a large number of false-negative results when a simple line propagation algorithm is used. This limits the application of this technique to only the core of prominent white matter tracts. By employing probabilistic path-generation algorithms, connectivity between a larger number of anatomical regions can be studied, but an increase in the number of false-positive results is inevitable. One of the causes of the inaccuracy is the complex axonal anatomy within a voxel; however, high-angular resolution (HAR) methods have been proposed to ameliorate this limitation. However, HAR data are relatively rare due to the long scan times required and the low signal-to-noise ratio. In this study, we tested a probabilistic path-finding method in which two anatomical regions with known connectivity were pre-defined and a path that maximized agreement with the DTI data was searched. To increase the accuracy of the trajectories, knowledge-based anatomical constraints were applied. The reconstruction protocols were tested using DTI data from 19 normal subjects to examine test-retest reproducibility and cross-subject variability. Fifty-two tracts were found to be reliably reconstructed using this approach, which can be viewed on our website.