INTRODUCTION: Diffusion MRI tractography is highly sensitive to modeling choices and tracking heuristics, yet objective validation is limited by the scarcity of microscopic ground truth aligned to MRI. We present a histology-informed tractography benchmark in post-mortem human corpus callosum (CC) white matter using diffusion MRI (dMRI) and co-registered proteolipid protein (PLP) myelin histology from the Oxford Digital Brain Bank.

AIM(S): We aim to introduce a histology-informed benchmark for tractography in post-mortem white matter in CC and test whether PLP-derived maps can be predicted from dMRI.

METHOD(S): We analysed three CC specimens and regions of interest (ROIs) in lateral/midline CC, corticospinal tract, and centrum semiovale. Using released MRI-histology registrations and PLP maps, we compared spherical deconvolution, tensor-based, FACT, and null-distribution tractography. Streamlines were mapped to histology and evaluated by tissue coverage and axial angular error vs local PLP orientations. We also predicted PLP orientation-dispersion from registered MRI orientation-dispersion and mean diffusivity using leave-one-sample-out regression and a small U-Net. Exploratively, we trained a conditional variational autoencoder (cVAE) to predict tissue-informed orientation distributions from MRI and derive tract-level anatomical constraint scores.

RESULTS: Tractography algorithm performance varied across specimens and methods. Null-distribution tractography gave control angular errors of ~45°. Selected methods showed lower angular error with moderate coverage. In the CC ROI, the U-Net showed moderate correspondence with PLP targets, whereas label-shuffled controls were near zero or negative. Exploratory cVAE analyses supported feasible PLP-informed orientation-distribution prediction and tract-level scoring on held-out data.

CONCLUSIONS: This work provides a reproducible microscopy-referenced benchmark for tractography in human CC white matter and a proof-of-principle MRI-to-histology prediction framework.

FINANCIAL SUPPORT: Horizon 2020 (Grant No. 857533), FNP (MAB PLUS/2019/13), and the Polish Ministry of Science and Higher Education (MEiN/2023/DIR/3796).