Rapid Patient-Specific Dose Estimation for Total Skin Electron Therapy Using AI-Derived Posture Models and GPU-Accelerated Monte Carlo
Abstract
Purpose
In Total Skin Electron Therapy (TSET) using the Stanford technique, the delivered dose is strongly affected by daily positioning uncertainties. This study presents an integrated framework that combines AI-assisted rapid generation of patient-specific six-posture models with GPU-accelerated Monte Carlo simulation to enable same-day cumulative dose estimation for TSET.
Methods
Patient postures were reconstructed using an AI-assisted human body modeling pipeline based on the SMPL framework. Silhouettes and joint keypoints were automatically extracted from clinical Cherenkov images using deep-learning segmentation and pose-estimation networks. A rapid optimization procedure estimated patient-specific body shape parameters and joint rotations, generating anatomically consistent finite-element meshes with identical vertex indexing across all six postures. These AI-generated models were then used as input to a GPU-accelerated Monte Carlo (PMC) engine employing EGSnrc pre-calculated electron tracks and a TOPAS-commissioned 6 MeV TSET beam phase space. Dose distributions were computed for each posture and summed on the surface mesh to obtain cumulative dose maps and dose surface histograms (DSHs).
Results
The AI framework successfully generated anatomically consistent six-posture patient models without requiring full 3D surface scans, reducing model preparation time from weeks to minutes. GPU-accelerated MC reduced simulation time from ~1 day per posture using CPU-based TOPAS to under 15 minutes per posture, while improving statistical uncertainty from ~10% to ~2% for clinically relevant surface doses. Cumulative dose distributions and DSH metrics showed strong agreement with TOPAS reference simulations on patient geometries.
Conclusion
By integrating AI-based rapid posture reconstruction with GPU-accelerated MC simulation, this work establishes a fast, accurate, and clinically feasible pipeline for patient-specific cumulative dose evaluation in TSET. The combined framework substantially reduces both modeling and computation time, enabling practical treatment assessment and supporting future real-time or adaptive TSET dose verification workflows.