Skip to content

Commit

Permalink
Update release_notes.md
Browse files Browse the repository at this point in the history
  • Loading branch information
villekf committed Jun 26, 2024
1 parent 9a760af commit 1d5c62e
Showing 1 changed file with 0 additions and 80 deletions.
80 changes: 0 additions & 80 deletions release_notes.md
Original file line number Diff line number Diff line change
@@ -1,85 +1,5 @@
# Release notes

## OMEGA v2.0.0

### New features

- Added two new projectors for CT data
- Projector type 4 is an interpolation-based GPU-only projector
- Projector type 5 is the branchless distance-driven projector (GPU only)

- Added support for SPECT reconstruction
- Supports also GATE projection data
- Collimator-detector response is only supported with implementation 2

- Added support for direction-vector based reconstruction in CT
- Added support for pitch/roll/yaw for cone-beam CT

- Added subset selection based on projection images/sinograms
- I.e. every Nth projection image/sinogram is selected
- Alternatively the projections can be selected randomly

- Added support for dual-layer PET reconstruction
- Supports both with and without detector offset
- Automatic GATE data import only supports dual-layer without detector offset

- Added ROOT support for Windows
- This uses "legacy" load that causes crashes on Linux systems, but works on Windows
- Requires 64-bit version of ROOT and thus at least Visual Studio 2022

- RDP and total variation is now faster to compute on implementation 2 and uses much less memory
- Does not apply to TV when using anatomical reference image

- Orthogonal distance-based and volume of intersection based projectors are faster to compute with implementation 2

- Removed MLEM as a separate algorithm
- MLEM can still be computed by selecting OSEM with 1 subset
- Any subset-based algorithm can now be run without subsets

- Only one algorithm/prior can be selected at a time with any implementation
- Previously implementation 2 allowed multiple different algorithms/priors

- Added five different image-based preconditioners and two measurement-based preconditioners
- For image-based these include diagonal, EM, IEM, momentum, normalized gradient and filtering based preconditioners
- For measuremented-based, only diagonal and filtering-based preconditioners are available
- Support is limited to only a few algorithms (MBSREM, PKMA, PDHG, FISTA)

- Added several new algorithms
- These include primal-dual hybrid gradient (PDHG), FISTA, LSQR, CGLS
- PDHG contains several different variations with and without regularization
- Works with any input data (PET, CT, SPECT)
- For CT, these use linearized data (this is done automatically)

- Added ramp-filtering with several different windowing functions
- Window functions include Hamming, Hann, Blackman, Nuttal, Parzen, cosine, Gaussian, and Shepp-Logan

- Added the ability to use mask images to limit LORs/measurements and/or voxels to take into account during reconstructions
- It is possible, for example, to only reconstruct a cylindrical region instead of the whole rectangular volume
- Alternatively, it is possible to take into account only measurements from a certain region
- Improves computation speed

- Added support for hybrid projectors
- Works with only a limited number of combinations, such as projector types 1 and 4
- Forward projector can thus be different from backprojection (this is already default with projector type 4)
- Projector type 4 can be used as a forward projector with PET data (setting `options.projector_type = 41`)
- The first value is the forward projector, while the second one is backprojection, i.e. 41 uses type 4 for forward and type 1 for backward

- For non-Poisson-based algorithms, positivity can be enforced with `options.enforcePositivity = true`

- Added support for object offset
- I.e. FOV does not need to be in the origin

- Combined all regularization parameters into one variable
- Rather than have separate regularization parameters for ALL function and prior combinations, only one `options.beta` is now used
- The change, however, is backwards compatible, i.e. you can keep using the old ones as well
- All new algorithms, however, only use the `options.beta` alue
- Should reduce clutter in the main-files

- The above change was also done for relaxation parameter lambda

- Removed support for PKMA sigma value
- Code still remains, although commented

## OMEGA v1.2.1

### Bug fixes and enhancements
Expand Down

0 comments on commit 1d5c62e

Please sign in to comment.