Dwidenoise in MS

Hi all,

I’d like to use the dwidenoise in MS population, and I’m worried about its performance in the presence of WM lesions. It has been used in MS before?

Thanks in advance,


Hi @eli82 ,

I’ve tested and checked the results myself now over a wide range of acquisition scenarios as well as varying WMHs (smaller and larger, both periventricular as well as deeper in the WM), and I’m pretty comfortable with the outcomes so far!
I’ve seen cases where the technique appears to behave slightly more conservative around those regions, compared to NAWM in the same subjects; which is in principle a good thing.
I haven’t come across cases yet where I would suspect the technique to be smoothing out any spatial features… but I do welcome your experiences as well. I suggest also outputting the noise map via the -noise option of dwidenoise; and sharing a screenshot or two if you’re concerned…

Hi Thijs,

Thanks for your quick answer and your help. We’ve uploaded the root data (001_DWI.tiff) and the output of noised map (001_DWI_noise_map.tiff). We’ve also attached the denoised image (001_DWI_denoise.tiff).

The DWI was a High Angular Resolution Diffusion Imaging (HARDI) sequence with TR/TE, 14800/103 ms; 100 contiguous axial slices; 1.5 mm isotropic voxel size; 154 x 154 matrix size; b value, 1000 s/mm2; 60 diffusion encoding directions and a single baseline image acquired at 0 s/mm2.




If you have any question, do not hesitate to contact us.

Best regards,


Hi Eli,

Those images definitely look fine; they’re also not too special or challenging in nature (I can’t spot any obvious major WMHs or other lesions; but then again, I’m not a trained MD :wink:). The noise map also looks like what I’ve come to expect, no worries there as well. The bigger challenge (for further CSD processing) will probably be the low b-value, although the denoising may actually help a great deal on that front! It definitely looks like the denoising did its job well.