A precise atlas of the human subcortex

Key results

1. 01

   ShownFOCUS delineates 85 bilateral and 10 midline subcortical grey and white matter structures (180 segmentations) at 100 µm isotropic resolution. It concentrates on deep brain stimulation (DBS) target zones, especially the mesencephalic-diencephalic transition zone.

   Drawn on the Edlow et al. 100 µm ex vivo 7T FLASH MRI of one postmortem brain. Boundaries were checked against 48 additional resources: histology with Gallyas, Nissl and H.E. stains, dark-field microscopy, fibre dissections and textbooks. Coverage includes basal ganglia laminae, thalamic nuclei, the Fields of Forel and the cerebellar nuclei (Fig. 1). · Abstract; Results; Manual segmentation of grey and white matter…; Figure legends

2. 02

   SuggestedA hand-refined nonlinear warp moves the 100 µm segmentations into MNI ICBM 2009b space at full resolution. It also yields an updated MNI version of the ex vivo template, which the authors report fits subcortical regions better than the original release.

   Five-stage ANTs registration (rigid, affine, whole-brain SyN, then two subcortex-focused SyN stages), refined with 62 paired anchor segmentations (31 per hemisphere) over 37 iterations plus local WarpDrive edits. Accuracy was judged by expert visual inspection and by matching to histological sections (Fig. 3). The comparison of original and refined templates is in Supplementary Materials, not in this text. · Results; Registration to MNI space using advanced…; Figure legends

3. 03

   SuggestedManual 'tract architectonics' with the new CurveToBundle tool produced 77 tracts. They include thin or tortuous pathways (the pallidohypothalamic tract, under 2 mm across; the fasciculus retroflexus; the oculomotor nerve) that the authors say diffusion tractography handles poorly.

   Streamlines were sampled along drawn curves (1000 for large cortico-cortical/cortico-subcortical tracts, 500 for major subcortical projections, 250 for smaller bundles), constrained by atlas structures. With segmentation guidance the ansa lenticularis matched fields H and H1, missed by an earlier 0.5 mm MNI reconstruction; an analogous comparison covers the fasciculus lenticularis (Fig. 5g-j). · Results; Manually curated Pathway Atlas; Figure legends

4. 04

   ShownIn a published subthalamic nucleus DBS fibre-filtering analysis, swapping the Netstim Atlas connectome for FOCUS tracts added subcortical tracts to the hyperdirect projections: indirect-pathway surrogates and substantia nigra, pedunculopontine and centromedian connections. Predictive performance did not change.

   Parkinson's disease: 129-patient discovery cohort (three centres) and 89-patient independent test cohort (two centres); outcome was improvement in motor UPDRS-III. Netstim Atlas v2.2: Spearman ρ = 0.33 circular (p = 0.0004), 0.29 ten-fold cross-validation (p = 0.001), 0.30 test (p = 0.006). FOCUS: ρ = 0.36 (p = 0.0002), 0.30 (p = 0.0008), 0.30 test (p = 0.003). · Results; DBS Fiber Filtering; Figure legends

5. 05

   SuggestedIn three dystonia patients after radiofrequency pallidothalamic tractotomy, atlas overlap separated the two responders (largest share in Forel field H1) from the non-responder (mostly zona incerta, no H1 overlap), even though the lesion centroids lay only 2.75-3.23 mm apart.

   Improvement on the Burke-Fahn-Marsden movement subscale (BFMDRS-M) was 89.8%, 66.7% and 0%. Share of lesion volume in H1: 32.8% and 38.7% in the responders. The non-responder's lesion was 61.1% zona incerta and 20.9% field H2. The region shared by both responders was 197 voxels, 55.8% of it within H1, and is crossed by the ansa and fasciculus lenticularis. · Results; Lesion study; Figure legends