Image reconstruction by event-based 3D TOF FBP

This application, written in R language, is devoted to image reconstruction for Jagiellonian PET (J-PET), using a combination of three independent kernel functions in image domain, similar to multivariate kernel density estimation (KDE), but with asymmetric definition.

Prerequisites

Tested on Ubuntu 16.04 LTE with R version 3.6.3 installed, as well as on Scientific Linux CERN SLC release 6.10, with R version 3.4.1. The script operates with some shell commands (for example, readlink -f <incomplete_path_to_file>), but. basically, all of them are standard.

Additional R packages are required:

• jsonlite
• Rcpp
• RcppArmadillo
• data.table (optional)
• nat (optional - for reading .nrrd sensitivity map)

There is no need to install anything. Just put directories modules/, source/ and cpp/ into a single location, in order to match the paths properly.

Input data and parameters

The application proceeds with GOJA output format (16-column ASCII) as the input, but uses only first 8 columns: (hit1_x, hit1_y, hit1_z, time_of_hit1, hit2_x, hit2_y, hit2_z, time_of_hit2). Cartesian coordinates hit*_* must be in centimetres, time tags -- in picoseconds.

The parameters for the reconstruction, sensitivity map generation, as well as the geometry of the scanner, are stored in JSON format: it is self-describing and easy to undersand. Please refer to the examples/ directory for the details.

Map formats for corrections

Attenuation and sensitivity corrections need external files with dedicated maps. By default, they are stored as .rds R format which contains R list of the following structure:

List of 4
 $ x        : num [1:223] -19.9 -19.7 -19.5 -19.3 -19.1 ...
 $ y        : num [1:223] -19.9 -19.7 -19.5 -19.3 -19.1 ...
 $ z        : num [1:191] -24.7 -24.5 -24.2 -24 -23.7 ...
 $ intensity: num [1:223, 1:223, 1:191] 0 0 0 0 0 0 0 0 0 0 ...

Alternatively, .nrrd can be used for sensitivity map (must be of the same dimensions as the reconstructed image), while plain ASCII is allowed for attenuation map, stored as follows:

"x"	"y"	"z"	"att.coeff"
-15.125	-11.725	-11.3	0
-14.925	-11.725	-11.3	0
...
15.075	-11.725	-11.3	0.096
-15.125	-11.525	-11.3	0.096
...
15.075	11.675	-11.3	0.096
-15.125	-11.725	-11.1	0.097
...
14.875	11.675	11.3	0
15.075	11.675	11.3	0

Here, a proper order should be preserved for al coordinates.

Output formats

Available output file formats are .rds and .nrrd. The former is native to R (has the same list structure as described above), while the latter is also used in J-PET MLEM application. You need external app to view .nrrd images or dedicated libraries, e.g. MRIcroGL, Aliza, pynrrd etc. Store the outcome as .rds only if you are familiar with basic R: you need to run it interactively or write a dedicated script to view the images.

Script architecture

The script is designed to work as quick as possible, hence R environments are utilised for encapsulation. The only exclusion has been made for JSON reader source/ReadFromJSON.R, which uses Reference class.

Usage

Assuming all script files are in the same place, execute simply the following to run the reconstruction:

$ cd <TOF_FBP_app_dir>
$ Rscript [--vanilla] source/LaunchReconstruction.R <path_to_json_params_file> [> log_file.log]

Here, <path_to_json_params_file> denotes the path to JSON file with parameters (see examples/). The option --vanilla prevents Rscript from reading R history, profile, or environment files, as well as reloading data or objects from previous sessions.