.Photoacoustic image simulation of synthetic ground truths

[tobychenkaka]

Hello everyone,

I am currently working on reproducing the results of the dataset https://doi.org/10.17863/CAM.96379, which involves generating 3D vascular structures using the Lindenmayer system (L-Systems) and performing photoacoustic image simulation using SIMPA and the k-Wave toolbox. I have already generated some complex vascular structures using the L-System and saved them in binary image format (or other required formats such as NIfTI).

Next, I aim to complete the following steps:

Import the generated vascular images into the SIMPA simulation pipeline.
Perform photoacoustic simulation on these vascular structures, including optical and acoustic forward modeling as well as image reconstruction.
I have gone through some of the example files provided by SIMPA (e.g., optical_and_acoustic_simulation.py), but I am unsure how to properly integrate the custom vascular structures into the SIMPA pipeline.

I would greatly appreciate your help, and if more context or code snippets are needed, please let me know!

[TomTomRixRix]

Sorry for the late reply. Happy to hear that you work on reproducing results from the community and use SIMPA for this. As far as I've understood, you have binary 3D segmentations of vascular structures as input and would like to perform the whole photoacoustic simulation pipeline (optical, acoustic and reconstruction). Am I right?

For this use case there is the SegmentationBasedAdapter to create virtual tissues based on your given segmentations. You can find an example of it's usage in simpa_examples/segmentation_loader.py. You basically have to define a segmentation volume mask (your binary vascular structures) and a segmentation class mapping. For the latter you would need to assign a class from the tissue library to each class in your segmentation so that optical and acoustic properties are automatically assigned to each voxel. As you mentioned that your data is binary, I would assume that there is one background class, e.g. muscle tissue sp.TISSUE_LIBRARY.muscle() and one vessel class, e.g. sp.TISSUE_LIBRARY.blood(). But you can also create new tissue classes or mix chromophores with the MolecularCompositionGenerator.

In case you already know the optical and acoustic properties for each voxel, it might be easier to construct and hdf5 file with the corresponding parameter maps as input and then only run the simulation pipeline without any tissue model creation.

To run the photoacoustic simulation pipeline you can find an example in optical_and_acoustic_simulation.py. Of course you might want to change some of the properties according to your needs.

I hope this help. If there are more detailed questions regarding the segmentation based adapter, maybe @kdreher or @faberno can assist further, as I don't have much experience with it myself.