OK, that sounds like a different problem again from the others mentioned in this thread…
My guess is your data come from a Philips scanner…? They have a habit of adding an additional image at the end of the series, which I think is a mean DWI image, computed from the acquired data. Clearly this is not an actual diffusion-weighted image, and typically its DW encoding entry looks ‘odd’ - on our scanners, it shows up as a 0 0 0 1000: a zero vector with a non-zero b-value… This is clearly going cause all manner of confusion for any downstream processing. You could verify this by inspecting the bvecs & bvals files to see if that is the case here (i.e. last column of bvecs contains zeros, but last entry of bvals is non-zero).
Assuming that is indeed the problem, you’ll probably find that updating your installation to the latest version, freshly released a few days ago, fixes the issue since it contains improved handling of DICOM data that should actually be able to deal with precisely these types of data (since the additional computed image shows up with a different ImageType entry in the DICOM headers)…