TG-43 comparison with egs_brachy simulations showing discrepancies near phantom edge

[vahx129]

Describe the bug I am not 100% confident this is a bug maybe so much as me trying to perform this validation incorrectly. But essentially, I'm trying to validate egs_brachy dose distributions in voxelized water phantom geometries by comparing them to TG-43 calculated doses from MIM which we use as our TPS. When I do the comparisons for post implant CTs from actual patient treatments, the ones where the seeds are relatively centrally placed in the axial plane of the CT image agree well with egs_brachy simulations. In the cases where the seeds are closer to the phantom edges, I'm seeing a lot more disagreement which I don't expect. It makes me think that maybe TG-43 scatter conditions aren't being faithfully reproduced even though I'm embedding my phantom geometry into a much larger box of water to approximate the infinite water conditions of TG-43.

To Reproduce A precise description of the steps to reproduce the behavior:

1. I convert a CT scan of a patient into an .egsphant geometry with a ramp that converts all densities to water and resize the voxels to be 0.5 mm in all directions.
2. I use a MATLAB script to place the Cs-131 seeds in the same positions as they were localized in the patient's post-implant scan and create an egs_brachy input file out of it that calls the .egsphant file (which is all water), embeds the seeds into their accurate treatment locations in that .egsphant file, and then embed the "seeds in egsphant" geometry into a very large box of water.
3. Run the simulation to create .3ddose files from egs_brachy and use a normalization to convert units to Gy based on mean lifetime and a TG-43 simulation of a single Cs-131 seed (cross-checked against the results from Carleton and came within about 0.5% of their normalization number). The seed strength in U is accounted for here as well since we only use uniform strength seeds when treating H&N patients.
4. Compare the 3ddose distribution in units of Gy to the DICOM RT-Dose file generated by the MIM TPS after seeds were localized in the post-implant CT scan.
5. Run various Gamma analyses to see that regions closer to the .egsphant edge have worse agreement with the TG-43 calculated doses than elsewhere.

Expected behavior I expect that my TG-43 TPS calculated dose distributions and Monte Carlo simulations, which are performed in a pure voxelized water phantom geometry itself embedded in a giant surrounding box of water to simulate infinite water conditions, should agree regardless of how close the seeds are to the edge of the voxelized portion of the phantom geometry. I can't say for certain that this isn't intended behavior based on how I'm running these simulations but when I look at the tracks in the "final" geometry of my simulations they look reasonable but since I'm only exporting the .3ddose of just the voxelized water phantom geometry to compare to my TPS dose, I'm wondering why the higher levels of disagreement are showing up so starkly near the phantom edge and not elsewhere. In cases where the seeds are centrally located within the phantom geometry, this doesn't seem to pop up at all.

Screenshots TPS Dose for one Axial Slice of clinical seed arrangement based on TG-43

egs_brachy dose for same axial slice with same clinical seed arrangement in a full water phantom

Global Gamma 2%/2mm dose comparison between TG-43 based calculation and egs_brachy in water phantom geometry calculation illustrating a lot of disagreement towards the edge of the phantom

egs_brachy tracks when visualizing just voxelized water phantom + seed geometry (same perspective as above images)

egs_brachy tracks when visualizing entire geometry which is seeds embedded in a voxelized water phantom matching resolution of CT scan itself embedded in a larger box of water

TPS Dose for a more centrally located set of seeds

egs_brachy dose for a more centrally located set of seeds

More favorable Gamma comparison between the previous two dose distributions

Operating system

• [X] Linux
• [ ] Windows
• [ ] macOS
• [ ] Other:

EGSnrc version egs_brachy 2024

Additional context I can share .egsphant, ramp, and input files if needed but am also still half of the mindset that I might just not be running these simulations properly. Either way, help would be hugely appreciated!

*Edited because I accidentally pasted the same screenshot twice of egs_brachy dose distribution instead of actual TPS one