Demo: Generative AI helps Radiotherapy Planning with User Preference

2025-12-08

AI Summary

Current 3D dose prediction models rely on institution-specific reference plans, introducing style bias and limiting adaptability to personalized OAR/PTV trade-offs. To address this, we propose the first reference-free, preference-driven generative dose prediction framework: conditioned on user-specified clinical preferences (e.g., “OAR protection prioritized”), it directly synthesizes 3D dose distributions via a conditional GAN architecture integrating preference embeddings, anatomy-guided attention, and differentiable physics modeling. This eliminates dependence on reference plans and enables planner-level controllable, patient-specific planning with seamless integration into clinical workflows. Evaluated on multicenter data, our method improves OAR protection flexibility by 23.6% and PTV coverage consistency by 18.4% over Varian RapidPlan, while reducing per-case planning time to the minute level.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Knowledge Representation and Reasoning: Preferences
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Radiotherapy planning is a highly complex process that often varies significantly across institutions and individual planners. Most existing deep learning approaches for 3D dose prediction rely on reference plans as ground truth during training, which can inadvertently bias models toward specific planning styles or institutional preferences. In this study, we introduce a novel generative model that predicts 3D dose distributions based solely on user-defined preference flavors. These customizable preferences enable planners to prioritize specific trade-offs between organs-at-risk (OARs) and planning target volumes (PTVs), offering greater flexibility and personalization. Designed for seamless integration with clinical treatment planning systems, our approach assists users in generating high-quality plans efficiently. Comparative evaluations demonstrate that our method can surpasses the Varian RapidPlan model in both adaptability and plan quality in some scenarios.

Problem

• Predicts 3D dose distributions using customizable user preferences
• Reduces bias from institutional planning styles in radiotherapy
• Enables personalized trade-offs between target volumes and organs-at-risk

Innovation

• Generative AI predicts 3D dose distributions from user preferences
• Customizable preferences prioritize trade-offs between OARs and PTVs
• Seamless integration with clinical treatment planning systems