Quantitative validation of in vivo tractography

Establish precise and reliable evaluation methods for quantitatively validating in vivo tractography results and comparing machine-learning tractography models.

Background

The in vivo experiments use visual inspection because no ground-truth scoring framework equivalent to the ISMRM phantom is available. Existing bundle-segmentation methods depend on parameters and stochastic procedures, and their scores may vary with the number and selection of bundles included in quality control.

The authors therefore identify in vivo validation as unresolved. A suitable evaluation method would provide reproducible quality metrics precise enough to compare tractography models across subjects and datasets, rather than merely support exploratory visual assessment.

References

In vivo validation remains an open problem.

— A foundation for systematic analysis of transformers and RNNs for tractography  (2610.01894 - Renauld et al., 1 Oct 2026) in Section 2.3, Experiment 5: Results in vivo; Section 4, Discussion, subsection “We still lack precise evaluation methods for in vivo data”