The DEBISim Simulation pipeline is illustrated as follows:
Data generation using DEBISim is done by running through each of the four operational blocks as shown in the figure. The entire pipeline can be run by calling data/dataset simulator scripts or through use of graphical user interfaces. The DEBISim simulation pipeline is highly modular and each block has dedicated modules to choose from different algorithms/methods/models for implementation in the pipeline. The blocks are explained as follows:
Reference Data Generator Block:
This block generates the reference volumetric image data as well as the metadata (See Shape Lists) for the X-ray projections are to be produced. Reference Data generation can be automated to create randomized datasets and can also be manually generated through user-interactive GUI or through use of baggage templates.
Forward Model Block:
The forward model block is responsible for setting up the geometric model of the X-ray scanner which determines the nature of X-ray projections generated. Basically, this block takes in the reference data image and creates the X-ray sinograms/projections as per the scanner geometry. DEBISim employs the ASTRA toolbox for generating templates for different types of scanners. This block is also responsible for assigning X-ray material properties according to the different material labels assigned in the reference data image.
Decomposer Block:
This block is used for processing Dual Energy CT projections generated from the forward model to extract the DECT cofficient line integrals such as Compton, PE coefficient line integral images. The block employs different decomposition methods to extract this data which can be reconstructed to obtain additional material based information such as the effective atomic number.
Reconstructor Block
This block creates the CT reconstructions from the simulated X-ray projections and obtains the attenuation/coefficient images as they would be generated from an actual scanner. The reconstructor block depends heavily on the scanner model chosen in the forward model block. These images which form the final output of the simulator are annotated by the reference data /metadata making them useful training data instances.
The entire pipeline can be executed through GUIs or by using the simulator to X-ray data/datasets. Further information on each of the blocks and the libraries can be found in the documentation.