Transformer model predicts 3D scatter fields for interventional suite radiation planning

A transformer-based model accurately predicted 3D x-ray scatter fields in a simulated interventional suite, supporting near real-time inference and substantial speedup over Monte Carlo simulation, aiming to enable rapid staff radiation protection planning.

• The model predicts air kerma per source photon, transported energy, and energy fluence via coordinate-conditioned regression at queried voxel locations.
• Training and testing used a single simulated suite with fixed geometry; inference time scaled with the number of queried voxels, enabling interactive whole-room evaluation on a GPU.
• A key limitation is that the model was developed and validated in only one simulated room geometry, so generalizability to other suites is unknown.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.