#28 · Tools & infrastructure

Spatial, connectivity, and harmonization constraints in brain imaging statistics

The dominant thread is treating spatial geometry, connectivity, or harmonization as explicit constraints in statistical brain imaging pipelines, though the four preprints differ in scope. The hippocampal tau preprint builds a principal surface for PET covariate analysis, combining two-stage inverse probability weighted regression with SuStaIn subtyping, and detects choroid plexus off-target contamination. TRAECR handles upstream PET preparation for statistical modeling through registration, MRI co-registration, and COMBAT/RAVEL harmonization. The TBI preprint pursues phenotype discovery across heterogeneous multi-site segmentations. JOSAConn is the only one of these that directly incorporates connectivity, adding dMRI tractography-derived maps to deep-learning cortical registration and reporting outperformance relative to FreeSurfer. New elements are contamination detection in tau PET and connectivity-informed cortical registration.