Technical Issue: Review and Integration of Open-Source Simulation and Dosimetry Tools

[devhliu]

Background

To support the proposed radionuclide tracer simulation platform, we need to identify, evaluate, and potentially integrate existing open-source tools focused on medical imaging simulation, organ segmentation, lesion inpainting, and voxel-based dosimetry.

Similar Techniques on GitHub (Partial List)

The following projects and issues were identified as relevant:

• EGSnrc: Monte Carlo simulation for dose calculation and imaging (see e.g. EGS GUI, Dose output for cylinders inscribed in water phantom).
• OpenGATE: Voxel dosimetry, PET/SPECT simulation (memory leak report, libtorch segmentation fault).
• VICTRE_MCGPU: GPU-based MC simulation for X-ray imaging (breast phantom software).
• matRad: Monte Carlo/analytical dose validation (validation script, proton plan error).
• MMC: Mesh-based Monte Carlo for photon transport (demo error).
• AutoPET II challenge: PET/CT lesion/organs segmentation using nnUNet (paper).
• SAMUS: Segment Anything Model for medical segmentation.
• SHARM: Segmented head anatomical models for dosimetry.

Note: Only 30 of 37 identified results are shown. For full results, see GitHub issue search.

Tasks

• Review functionality and APIs of the listed projects.
• Benchmark key modules (segmentation, Monte Carlo, dosimetry, TAC simulation).
• Propose integration architecture or workflow for the new platform.
• Document technical gaps and requirements for new development.

Related Techniques (from Issues)

• Voxel-based dosimetry: memory leak (OpenGATE), out-of-bounds indexing (EGSnrc), validation scripts (matRad).
• Organ/lesion segmentation: U-net/nnUNet, SHARM, SAMUS.
• Monte Carlo photon/electron transport: MMC, EGSnrc, OpenGATE.

Next Steps

• Prioritize tools for prototype integration.
• Evaluate compatibility with PyVoxelDosimetry data models.
• Create development tasks for missing features or technical improvements.