v1.6

Version 1.6 release notes

1. Speed improvements to backprojection algorithms where the voxel size is "large" (splatting the voxel onto the detector is bigger than 2 detector pixels. Speed improvement ranges from 10% to 70%, depending on the CT geometry and CT volume parameters.
2. Enabled 3x5x5 median filter (before could just do 3x3x3 or 2D median filters).
3. Added lots of comments to C++ and CUDA header files.
4. Fixed bug in filtering step of helical FBP that would appear occasionally (program would crash) for large projection data sets.
5. As usual, updated the LEAPCT readthedocs page.

v1.5

Version 1.5 release notes

This release added some new denoising algorithms and features.

1. Added the ability to change the number of neighbors used in TV. Currently can choose 26 or 6. See the set_numTVneighbors and get_numTVneighbors functions.
2. Added (Scaled) Bilateral Filter denoising algorithm. See new demo script for usage example.
3. Added Dictionary Denoising algorithm (i.e., sparse representation of an overcomplete dictionary). See new demo script for usage example.
4. As usual, updated the LEAPCT readthedocs page.

v1.4

Version 1.4 release notes

1. Added two-material beam hardening correction (BHC) algorithm. If there is interested from the users, I can extend this to an arbitrary number of materials in a future release.
2. Changes to the API for physics-based correction algorithms, such as BHC and dual energy decomposition. These changes came from a new release of the XrayPhysics package which made changes and resolved some Linux build issues.
3. Fixed minor bug in calculation of TV functional scalar value
4. Indexing for all noise filters, including TV now uses uint64, so there is no limit on the array size other than RAM. Previously large arrays were processed in smaller chunks so that the number of elements did not exceed 2^31. The biggest impact this will have is now one can do GPU-GPU operations of large arrays and avoid CPU-GPU data transfers.
5. Enabled 2D ramp filter to operate on data on the GPU (in-place). Also made very significant speed improvements to the algorithm.
6. Added 2D ramp-based and SARR preconditioners to the RWLS algorithm. The SARR preconditioner turns RWLS iterations into a similar step as Iterative Filtered Backprojection (IFBP).
7. Added leapct.print_cost variable. If this is set to true, algorithms will print out the cost at every iteration.
8. Bug fixes

v1.3

Version 1.3 release notes

1. More than 4X helical FBP speed improvement
2. Implemented helical FBP optimized for large voxels
3. Fixed some bugs in RDLS
4. Reduced memory requirements for RWLS and RDLS
5. Fixed bug in modular-beam backprojector (axially-aligned worked fine, only really custom geometries had demonstrated occasional issues)

v1.2

Version 1.2 release notes

1. Added the option to specify a mask on the projection data for all iterative reconstruction algorithms. Note that no explicit mask is given for WLS or RWLS because one can just use the weighting matrix, W, as a mask.
2. Added dual energy decomposition and the ability to convert to effective atomic number (Ze) and electron density volumes (rhoe). Now LEAP can do basic SIRZ processing, though some minor features are still missing. These will be added soon.
3. Fix the scaling of the FBP algorithms for symmetric objects when the voxel sizes are different than their default value.

v1.1

Version 1.1 release notes

1. We sincerely apologize for changing the definition of a parameter, but we needed to redefine how z-slices were positioned which effects offsetZ. The documentation has been changed to reflect the change. Basically, the old method was causing confusion to myself and others and I think the new definition is much more intuitive.
2. Bug fixes!
3. Ramp filter speed improvement.
4. 5X speed improvement to backprojectors with large voxels
5. Added preprocessing algorithms: make attenuation radiographs, outlier removal, detector deblur, ring removal, beam hardening correction, find centerCol
6. Added more demo scripts to demo_leapctype folder
7. Added inconsistency reconstruction algorithm
8. enabled iterative reconstruction algorithm with pytorch tensors

v1.0

First official release of LEAP.