I’m adding to this thread, because I too am working with rodent DWI and have some interesting results from the dwidenoise command. This is my first time using dwidenoise, and none of the results I got seem to fulfill the “good residual” requirement of being free of anatomy.
Attached image shows the noise map and rms residual for dwidenoise, using three different settings for extent: 3, 5, and 7. Initially, I ran dwidenoise with default extent, yielding the middle image. It seemed suspicious, since there was definitely a brain outline. After consulting with others, I ran using the other extent setting, resulting in the other two results, which curiously shows even more anatomical structure.
This dataset contains 107 volumes, with 10 b0 images, and is obtained ex-vivo acquired over the course of 12 hours. Is this perhaps why so little noise is estimated with the default setting? This is one of the cleaner data, is this good data actually getting in the way of dwidenoise?