We've been looking at your data and found a problem with the way Siemens has written out the slice positions. Your data apparently is RAS or LAS (we can't tell for sure), but to3d determines the data to be RAI. The slice positions are increasing, and based upon DICOM standards, this would normally indicate inferior to superior. Your data, however, is stored superior to inferior; hence, the data is upside down.

Creating AFNI bricks will use the non-oblique transformation but will determine the orientation one way while creating NIFTI datasets will determine the transformation but still relying on a transformation matrix from correct slice position (using at least the direction of the slices).

To get around this problem, you can enforce an orientation with -orient RAS on the to3d command line for AFNI format datasets. You should examine the dataset to see if the bounding box and origin match what you expect from the acquisition parameters. It's impossible for us to know this bit of information given the incorrect definitions in the DICOM header, so we'll have to rely on the image acquistion parameters recorded on the scanner. To create NIFTI datasets, you'll have to remove the incorrect transformation matrix with 3drefit -deoblique or correct it, possibly by negating the the third row of the s-form transformation matrix using nifti_tool.

Deobliquing or obliquidating the data requires a cardinal dataset to know whether the transformation is correct. You can either upload a cardinal anatomical dataset from the same scanning session, or you can experiment on your own with various combinations of 3drefit to change the orientation, 3dresample to change slice order and 3dWarp to apply the transformation.