I noticed in my data that residuals of most voxels from dwidenoise are less than 1. But the raw value of these voxels are larger than 200. Does this mean dwidenosie reduce little noise?
or, should we use a signal change index(res/raw) to check if dwidenoise work well in the data?
If you say that the raw voxels of these voxels are larger than 200, do you mean in the b=0 shell, or at what b-value? An SNR ≈ 200 at b=0 may sound like a lot, but if you assume an average ADC ≈ 1.0um²/ms it leads to SNR ≈ 10 at b=3000s/mm² (using a very simple first-order approximation SNR(b) = SNR(0) . exp(-b.ADC) ).
If you want, you can check the SNR per shell:
dwishellmath dwi.mif mean bmean.mif mrcalc bmean.mif noiselevel.mif -divide snr.mif mrstats snr.mif -mask mask.mif
I would not rely on SNR (or 1/SNR as you suggested) for quality assurance. Just check the residuals for anatomical structure as recommended in the docs.
For 200, what i mean is the dwi image intensity with b=2000.
Thanks for your suggestion. I’ll try them.