Probabilistic Mapping and Automated Segmentation of Human Brainstem White Matter Bundles

Brainstem white matter bundles are essential conduits for neural signals that modulate human consciousness and homeostasis. Yet, their small size and complex architecture make segmentation challenging in neuroimaging studies. We leveraged diffusion MRI tractography to create the BrainStem Bundle Tool (BSBT), which automatically segments eight brainstem white matter bundles. BSBT performs unsupervised segmentation with a convolutional neural network on a custom probabilistic fiber map generated from tractography. Validation on healthy subject in vivo diffusion MRI, ex vivo diffusion MRI with histology, and test-retest analyses show BSBT's robustness across diffusion MRI acquisition protocols. We demonstrate the generalizability of BSBT through discrimination tasks in Alzheimer's disease, Parkinson's disease, and traumatic brain injury cohorts. Finally, we explore the prognostic utility of BSBT in a longitudinal analysis of a comatose patient with a midbrain lesion. BSBT creates opportunities to map brainstem white matter and investigate its role in a broad spectrum of neurological disorders.

-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Included in this dataset are: The **raw diffusion MRI** volumes for each brain specimen, the corresponding **ultra-high resolution FLASH** sequences (200um isotropic spatial resolution for C1/C2/EXC007/EXC012/EXC020/EXC021 and 550um isotropic spatial resolution for C4), and the representative reconstructed **Probabilistic Fiber Maps** (PFM) in the brainstem. Manual annotations and corresponding BSBT segmentations for all HCP, ADNI3 and ex vivo cases used for validation are included in the derivatives folder. Original HCP and ADNI3 dMRI scans on which this validation was performed on (with the same subject IDs can be accessed at https://db.humanconnectome.org and https://adni.loni.usc.edu respectively.

In the main manuscript, each brain specimen is referenced as follows:

sub-C1: **S1** <br>
sub-C2: **S2** <br>
sub-C4: **S3** <br>
sub-EXC007: **S4** <br>
sub-EXC012: **S5** <br>
sub-EXC020: **S6** <br>
sub-EXC021: **S7** <br>