New kappa #36
Merged
New kappa #36
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Member
Author
|
For Siemens-Vision600_ZrNEMAIQ kappa (slice 80) before: Note that the central value is changing quite a lot. |
For very noisy data, H.1 becomes quite noisy itself, with large impact on the kappa, and therefore a prior that depends on the noise in the data. The approximation we use now is `kappa^2 = A^T . ((1 / (A.x + c)) * A.1 )` i.e. the "central part" of the Hessian `y/(A.x+c)^2` is replaced. Moreover, we first filter the initial image with a 3D Gaussian with the given FWHM. WARNING: this makes the kappa calculation specific to Poisson obj funcs. WARNING: compute_kappa_image now takes an acq_model as first argument, no longer the obj_fun
For noisy data, H_loglikelihood.1 can be noisy itself. It's probably a bad idea to use a noisy preconditioner, so now using the (not noisy) kappas.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Avoid issues with kappa being quite noisy for very low-count data. in particular, kappa can be quite low for certain voxels, causing underregularisation. Moreover, it is strange to use a prior that depends on the noise in the data.
The approximation we use in this PR is
$\kappa^2 = A^T . ({1 \over A.x + c} * A.1 )$ $y/(A.x+c)^2$ is replaced.
i.e. the "central part" of the Hessian
Moreover, we first filter the initial image with a 3D Gaussian with the given FWHM (defaulting to 8mm).
WARNING: this makes the kappa calculation specific to Poisson obj funcs.
WARNING:
compute_kappa_imagenow takes anacq_modelas first argument, no longer theobj_funOther modifications:
create_initial_imagesnow firstchdirinto<data_path>, so no longer needs to be run in the directory with the data.kappa^2as preconditionerNote: currently on top of #29