Rick,

We created these datasets with the intention of having them aligned properly, but we're not sure we succeeded in doing that. One of the datasets is created from our data and given 4x4x4 voxels with adwarp; the other is a ROI mask created by resampling an atlas to 4x4x4 resolution with 3dresample -dxyz 4 4 4. I tried redoing the mask by using -master and it gave me a dataset that agrees with our original mask in all respects except that IJK_TO_DICOM agrees with the master and of course IDCODE_STRING, IDCODE_DATE and HISTORY_NOTE have new values. The .BRIK files do not differ at all. What I do not understand is how the IJK_TO_DICOM conversion matrices can differ when the voxel dimensions, orientations and dataset origins do not. Here's the information from the header, in a slightly more human-readable form:

common attributes:

ORIENT_SPECIFIC:    1   2   4   (LPI)

ORIGIN: 88  124 -70

DELTA:  -4  -4  4

IJK_TO_DICOM_REAL:
    -4  0   0   88
    0   -4  0   124
    0   0   4   -70

The 'good' IJK_TO_DICOM matrix is identical to the IJK_TO_DICOM_REAL matrix above, but the 'broken' one is

IJK_TO_DICOM:
    -2  0   0   90
    0   -2  0   126
    0   0   2   -72

The first matrix clearly agrees with the rest of the information, but I can't tell where the second one comes from or why it wants to dilate the dataset by a factor of 2.

Isaac