Quality of brainstem ROI registration using structural and diffusion MRI

• Incorporating diffusion MR images improved ROI registration accuracy for brainstem nuclei (red nucleus, substantia nigra) versus T1-only pipelines in n=20 healthy participants (Frontiers in Neuroscience).¹
• Improves reliability of BOLD/ROI sampling in brainstem for stimulation targets and fMRI-based interfaces.^{1 1}
• Authors are Yi-An A. Chen, Lars Kasper, Clement Chow, Yu Kuo, Alexandre Boutet, Jürgen Germann, Andres M. Lozano, Kâmil Uludağ, Andreea O. Diaconescu, and Sriranga Kashyap from the Krembil Brain Institute / University Health Network and University of Toronto (published 27 February 2026).¹
• The study evaluated four registration pipelines: T1w-only (baseline), FA-based, b0-based, and a multivariate combination, applied to three target nuclei: red nucleus (RN), substantia nigra (SN), and dorsal raphe nucleus (DRN).¹
• Data were acquired on a Siemens Magnetom Prisma 3 T scanner; DWI comprised 64 directions at b=2000 s/mm² plus 30 directions at b=1000 s/mm² in 20 healthy participants (10 female, mean age 42.15 ± 15.71 years).¹
• For the red nucleus, all diffusion-augmented pipelines significantly outperformed T1w-only (one-way ANOVA: F(3,57)=61.31, p<0.001); multivariate pipeline produced the highest Dice coefficients (t(19)=−11.44 vs. T1w, p<0.001).¹
• For the dorsal raphe nucleus, FA-based compound transforms produced the lowest mis-registration fraction; FA vs. T1w comparison: t(19)=8.82, p<0.001; the multivariate approach also significantly outperformed T1w (t(19)=4.55, p=0.0002).¹
• Jacobian determinant gradient maps revealed that T1w-based transforms emphasized the brainstem surface, while FA- and b0-based maps emphasized internal brainstem features, which are more relevant to nuclei localization.¹
• Clinical implication: inaccurate brainstem ROI registration risks unreliable BOLD signal sampling from deep nuclei; diffusion-augmented pipelines may improve robustness of fMRI studies for brainstem disorders (e.g., Parkinson’s disease) where disease-specific templates are advisable.¹
• The Brainstem Navigator Atlas v0.9 contained 68.1% of DRN voxels and 51.2% of RN voxels within the study’s probabilistic ROI boundaries, indicating systematic atlas-to-subject registration variability that diffusion pipelines partially mitigate.¹

Footnotes

1. https://www.frontiersin.org/articles/10.3389/fnins.2026.1711863 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹