Hi all,
I am transforming Allen Human Brain Atlas tissue sample locations for donors H0351.2001 and H0351.2002 into MNI152NLin2009bAsym space and wanted to check whether my coordinate-transform chain is valid.
The AHBA SampleAnnot.csv file provides mri_voxel_x/y/z indices for each sample. My current workflow is:
-
Identify STN samples by ontology label and well ID.
-
Treat
mri_voxel_x/y/zas zero-based voxel indices in the donor MRI voxel grid. -
Convert voxel indices to donor-native physical coordinates using the affine matrix from the corresponding donor T1w NIfTI header.
-
Convert coordinates from NIfTI/SPM RAS convention to ITK LPS convention before using
antsApplyTransformsToPoints. -
Apply the donor-specific nonlinear deformation field from donor-native space to MNI152NLin2009bAsym space.
-
Convert transformed coordinates back to RAS for reporting.
-
QC this by creating uniquely labelled 2 mm marker images around each donor-native sample coordinate, transforming these marker images to MNI space with nearest-neighbour interpolation, and comparing the resulting marker centroid to the directly transformed point.
Direct-point and marker-centroid estimates differ by <0.5 mm, and hemispheric laterality is preserved.
My question is: does this look like a valid coordinate transform chain for AHBA voxel-index sample coordinates, especially regarding zero-based voxel indices, RAS↔LPS conversion, transform direction/order, and the marker-centroid QC approach?
Many thanks!
