Notes:
This update fixes issues with the way noise is added to the images.
In the previous version, the default noise_level (0.06) was added
to the clean images assuming their intensities were standardised to
1. However, using the default parameters in v1.0 and below, the
intensity in the b=0 images is 8000, causing the added noise to be
8000 times to low. This has been fixed in this release by
standardising the clean images to 1. To add noise to previously
generated images please scale the noise to an appropriate level
(i.e. noise_level = 480).
There is also an issue in v1.0 and below when noise is added to
small datasets. In this case, images simulated from consecutive
gradient directions may end up with the same noise realisation as
the random number generator is seeded by the CPU clock at the start
of each image and the process may run faster than the time it takes
to iterate the clock. This has been fixed by only seeding the
random number generator only once at the start of the 'noisify'
tool.
Changes:
- Changed 'mri_sim' to produce image intensities standardised to 1
(i.e. b=0 image intensities are 1)
- Only seed the rng once in 'noisify' to avoid two images using the
same noise realisation, which occurred if the CPU clock hadn't
updated between the processing of two images.
|