Optimized for Intel multi-core processors
Source code available at:
https://github.com/HPImaging/sv-mbirct
Official Python interface for this package:
https://github.com/cabouman/svmbir
https://svmbir.readthedocs.io
Demo scripts and data files for running this program are available
here and also under a separate repository:
https://github.com/HPImaging/mbir-demos
Further references on MBIR and the technology in the high-performance implementation of this
code can be found at the bottom of this page, in the documentation accompanying the OpenMBIR
project and on Charles Bouman's website:
https://github.com/cabouman/OpenMBIR-ParBeam
http://engineering.purdue.edu/~bouman/publications/pub_tomography.html
http://engineering.purdue.edu/~bouman/publications/pub_security.html
http://engineering.purdue.edu/~bouman/publications/pdf/MBIP-book.pdf
--OR--
From a shell prompt, cd into the src folder and run make. Use one of the following forms depending on your available compiler:
make CC=icc # Intel compiler (fastest)
make CC=gcc # GNU compiler
make CC=clang # Apple/MacOS C compilerIf compiling is successful the binary mbir_ct will be created and moved into the bin folder.
ICC Tip: Initially after installing Intel Parallel Studio XE, there may be complaints of missing libraries when linking and running the code. Most issues can be resolved by executing the following line, which should be included in your .profile (or .bashrc, or whatever relevant resource file is executed each time a shell is launched).
source /opt/intel/bin/compilervars.sh intel64The path above assumes the install point for Parallel Studio XE was "/opt", which is the default for Linux and macOS. If not, you may need to poke around to find the location of "compilervars.sh".
To print a usage statement:
./mbir_ct -helpThe program is able to run completely with a single command call, but it's usually preferrable to run the reconstruction in two stages. In the first stage, the sytem matrix is precomputed and stored, and the second stage performs the actual reconstruction. Both stages use the executable mbir_ct. The system matrix can take a significant time to compute, however the matrix is fixed for a given geometry and data/image dimensions, so the matrix file can be reused for any scan that uses the same sinogram and image parameters.
(Note the accompanying demo scripts include a utility that detects whether the necessary sytem matrix file has already been computed and is available, given the input image/sino parameters, and the script automatically reads the file if available, or computes/stores it if not.)
./mbir_ct
-i <basename>[.imgparams] : Input image parameters
-j <basename>[.sinoparams] : Input sinogram parameters
-m <basename>[.2Dsvmatrix] : Output matrix fileIn the above arguments, the exensions given in the '[]' symbols must be part of the file names but should be omitted from the command line. Further description of data/image filenames is provided further down.
Example: (written as if file names have been assigned to variables in a shell script)
./mbir_ct -i $parName -j $parName -m $matName./mbir_ct
-i <basename>[.imgparams] : Input image parameters
-j <basename>[.sinoparams] : Input sinogram parameters
-k <basename>[.reconparams] : Input reconstruction parameters
-s <basename>[_sliceNNN.2Dsinodata] : Input sinogram projection file(s)
-r <basename>[_sliceNNN.2Dimgdata] : Output reconstructed image file(s)
(following are optional)
-m <basename>[.2Dsvmatrix] : INPUT matrix (params must match!)
-w <basename>[_sliceNNN.2Dweightdata] : Input sinogram weight file(s)
-t <basename>[_sliceNNN.2Dimgdata] : Input initial condition image(s)
-e <basename>[_sliceNNN.2Dprojection] : Input projection of init. cond.
-f <basename>[_sliceNNN.2Dprojection] : Output projection of final image
-p <basename>[_sliceNNN.2Dimgdata] : Proximal map for Plug & Play
: -p will apply proximal prior
: generally use with -t -e -fNote the default prior model is a q-QGGMRF with a 10-pt 3D neighborhood.
Example: Compute reconstruction
./mbir_ct -i $parName -j $parName -k $parName -s $sinoName \
-w $wgtNname -r $recName -m $matNameExample: Compute projection of input image only
./mbir_ct -i $parName -j $parName -m $matName \
-t <basename>[_sliceNNN.2Dimgdata] -f <basename>[_sliceNNN.2Dprojection]In the above examples, if -m is omitted the system matrix will be computed prior to starting the reconstruction (or projection).
The program can perform the inversion step for Plug & Play MBIR. The argument -t specifies the initial image state, and the -r argument specifies the output.
The following example uses the same file names for the input (initial state) and the output. These file names can be different if you want to save the intermediate image state in each ADMM iteration.
./mbir_ct -i $parName -j $parName -k $parName -s $sinoName \
-w $wgtName -m $matName -p $proxmapName -t $imgName -r $imgNameThe following form also reads in the projection (-e) of the input image state, and writes out the projection of the final image state (-f) so that the projection doesn't have to be re-computed in each mbir_ct call.
./mbir_ct -i $parName -j $parName -k $parName -s $sinoName \
-w $wgtName -m $matName -p $proxmapName -t $imgName -r $imgName \
-e $projName -f $projNameFor a detailed description of the contents and format for all the data and parameter files used in this program, see the documentation in the OpenMBIR project referenced at the top of this readme, directly linked to here. Also see the demos for specific examples.
The following parameter files are required, all in simple text:
<basename>.sinoparams
<basename>.imgparams
<basename>.reconparams
<view_angles_file.txt>Note these show the same generic basename but the names of all the input files are independent as they're specified in different arguments in the command line.
For the files containing sinogram or image data, the associated 3D data is split across files, one file per slice. The naming convention for the different files is as follows:
<basename>_sliceNNN.2Dimgdata
<basename>_sliceNNN.2Dsinodata
<basename>_sliceNNN.2Dweightdata
<basename>_sliceNNN.2Dprojectionwhere "NNN" is a slice index. The slice indices must be a non-negative integer sequence, where the indices include leading zeros and no spaces (e.g. 0000 to 0513). The number of digits used for the slice indices is flexible (up to 4) but must be consistent across all the files used in a given reconstruction call.