|
Notes:
This version fixes issues with the noisy images inline with changes
made in v1.0.1 of the software. The default 'noise level' (standard
deviation of Rician noise) in v1.0 was 4 orders of magnitude too
low as it was added assuming that the b=0 images in the clean data
set were normalised to 1 (in reality they were 8000). In addition,
consecutive gradient direction images may have received the same
noise realisation as the noise was seeded of the CPU clock for
every direction and it may not have had time to iterate between
each gradient direction. This has now been fixed
Changes:
- Clean b=0 images have been normalised to 1.
- SNR is now a much more realistic 16.7
- Noise realisations are now guaranteed to be different between
consecutive gradient directions.
|
