Bundle analytics, a computational framework for investigating the shapes and profiles of brain pathways across populations

Abstract

Tractography has created new horizons for researchers to study brain connectivity in vivo. However, tractography is an advanced and challenging method that has not been used so far for medical data analysis at a large scale in comparison to other traditional brain imaging methods. This work allows tractography to be used for large scale and high-quality medical analytics. BUndle ANalytics (BUAN) is a fast, robust, and flexible computational framework for real-world tractometric studies. BUAN combines tractography and anatomical information to analyze the challenging datasets and identifies significant group differences in specific locations of the white matter bundles. Additionally, BUAN takes the shape of the bundles into consideration for the analysis. BUAN compares the shapes of the bundles using a metric called bundle adjacency which calculates shape similarity between two given bundles. BUAN builds networks of bundle shape similarities that can be paramount for automating quality control. BUAN is freely available in DIPY. Results are presented using publicly available Parkinson’s Progression Markers Initiative data.

Neurobiological and behavioural outcomes of biofeedback-based training in autism: a randomized controlled trial

“…Additionally, participant head motion during DWI acquisition was quantified using the root mean square movement summary from eddy. 37 Delineation of the bilateral CST on the DWI images was done using TractSeg, a tool that utilizes a pre-trained convolutional neural network to create bundle-specific tractograms 38 from which we calculated mean FA across the entire CST as well as across 20 equally distanced segments along the streamlines 39 ; https://github.com/MIC-DKFZ/TractSeg/ . See Supplementary methods for more detail.…”

“…After the null results of the primary analysis, follow-up analyses divided the tract into 20 distinct segments (from superior to inferior) and also delineated the left from the right tract. 39 Follow-up analyses of the CST used fdr to control for multiple comparisons. See Supplementary methods for more detail.…”

Cortically constrained shape recognition: Automated white matter tract segmentation validated in the pediatric brain

“…This can be improved by iterating on the local registration of bundles using Streamline Linear Registration and re-running the RecoBundles algorithm on the locally aligned bundle template, or potentially with a recently published step added to RecoBundles called Auto-calibration. 54 Another area to explore further is tuning the RecoBundles parameters in a tract-and cohort-specific capacity. Although this study was designed to evaluate the performance of this approach using parameters uniformly, one of the main potential improvements to the performance is investing effort in customizing the pipeline to the cohort by tuning the parameters for each bundle and selecting (or creating) a complementary bundle atlas.…”

“…The use of multiple centroids could potentially improve the performance of CCSR especially in anatomically complex tracts, as it has been recently suggested. 54 Another avenue of improvement is developing a custom template using bundles more specific to this cohort (e.g., acquisition, tractography algorithm, and pediatric healthy controls). Custom templates can be created with a single representative case (not preferred due to single-subject bias), from a group cohort with q-space diffeomorphic reconstruction, 55 or by combining the bundle shapes from a cohort of manually segmented cases if a large and heterogeneous set of segmented bundles is available (code provided).…”

“…One challenge in using shape‐based methods is that streamline registrations are inherently linear, so a global whole‐brain registration between dissimilar brains may result in significant mismatches in positioning of tracts for certain parts of the brain. This can be improved by iterating on the local registration of bundles using Streamline Linear Registration and re‐running the RecoBundles algorithm on the locally aligned bundle template, or potentially with a recently published step added to RecoBundles called Auto‐calibration 54 . Another area to explore further is tuning the RecoBundles parameters in a tract‐ and cohort‐specific capacity.…”