Surrogate Modelling of Proton Dose with Monte Carlo Dropout Uncertainty Quantification

2025-09-16

AI Summary

Monte Carlo (MC) dose calculation in proton therapy offers high accuracy but is computationally prohibitive for clinical workflows requiring frequent re-execution—such as robust optimization and adaptive replanning. To address this, we propose a differentiable neural surrogate model that integrates Monte Carlo Dropout to enable voxel-wise dose prediction alongside rigorous uncertainty quantification. Crucially, our method decouples epistemic uncertainty (model confidence) from aleatoric uncertainty arising from input distribution shift, enabling accurate attribution of uncertainty sources—particularly at material interfaces and under distributional shifts. The model is validated against analytical 1D solutions, 2D bone–water phantoms, and 3D water phantoms using high-fidelity TOPAS/Geant4 simulation data. It preserves MC-level accuracy while achieving a 100–1,000× speedup, thereby enabling uncertainty-aware robust treatment planning and real-time adaptive optimization.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Stochastic Optimization
• Planning, Routing, and Scheduling: Scheduling under Uncertainty

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Accurate proton dose calculation using Monte Carlo (MC) is computationally demanding in workflows like robust optimisation, adaptive replanning, and probabilistic inference, which require repeated evaluations. To address this, we develop a neural surrogate that integrates Monte Carlo dropout to provide fast, differentiable dose predictions along with voxelwise predictive uncertainty. The method is validated through a series of experiments, starting with a one-dimensional analytic benchmark that establishes accuracy, convergence, and variance decomposition. Two-dimensional bone-water phantoms, generated using TOPAS Geant4, demonstrate the method's behavior under domain heterogeneity and beam uncertainty, while a three-dimensional water phantom confirms scalability for volumetric dose prediction. Across these settings, we separate epistemic (model) from parametric (input) contributions, showing that epistemic variance increases under distribution shift, while parametric variance dominates at material boundaries. The approach achieves significant speedups over MC while retaining uncertainty information, making it suitable for integration into robust planning, adaptive workflows, and uncertainty-aware optimisation in proton therapy.

Problem

• Accelerating proton dose calculations for clinical workflows requiring repeated evaluations
• Providing fast differentiable dose predictions with voxelwise uncertainty quantification
• Separating epistemic and parametric uncertainty contributions under distribution shifts

Innovation

• Neural surrogate with Monte Carlo dropout
• Fast differentiable dose predictions with uncertainty
• Separates epistemic and parametric variance contributions