FluenceFormer: Transformer-Driven Multi-Beam Fluence Map Regression for Radiotherapy Planning

2025-12-26

AI Summary

In radiotherapy planning, inferring multi-beam fluence maps from dose distributions is an ill-posed inverse problem; existing CNN-based methods struggle to capture long-range anatomical-geometric dependencies, often yielding structurally distorted or physically infeasible solutions. This paper proposes a two-stage Transformer framework: the first stage generates anatomy-guided global dose priors, while the second stage fuses beam geometric information to regress physically realizable fluence maps. We introduce Fluence-Aware Regression (FAR) loss—a novel objective unifying voxel-wise accuracy, gradient smoothness, structural consistency, and beam energy conservation. The architecture is backbone-agnostic, compatible with medical Transformers such as Swin UNETR. Evaluated on a prostate IMRT dataset, our method reduces energy error to 4.5% and achieves significantly higher structural fidelity than both CNN-based and single-stage approaches (p < 0.05).

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Multi-modal Vision
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization

Abstract

Fluence map prediction is central to automated radiotherapy planning but remains an ill-posed inverse problem due to the complex relationship between volumetric anatomy and beam-intensity modulation. Convolutional methods in prior work often struggle to capture long-range dependencies, which can lead to structurally inconsistent or physically unrealizable plans. We introduce extbf{FluenceFormer}, a backbone-agnostic transformer framework for direct, geometry-aware fluence regression. The model uses a unified two-stage design: Stage~1 predicts a global dose prior from anatomical inputs, and Stage~2 conditions this prior on explicit beam geometry to regress physically calibrated fluence maps. Central to the approach is the extbf{Fluence-Aware Regression (FAR)} loss, a physics-informed objective that integrates voxel-level fidelity, gradient smoothness, structural consistency, and beam-wise energy conservation. We evaluate the generality of the framework across multiple transformer backbones, including Swin UNETR, UNETR, nnFormer, and MedFormer, using a prostate IMRT dataset. FluenceFormer with Swin UNETR achieves the strongest performance among the evaluated models and improves over existing benchmark CNN and single-stage methods, reducing Energy Error to $mathbf{4.5%}$ and yielding statistically significant gains in structural fidelity ($p < 0.05$).

Problem

• Predicts fluence maps for automated radiotherapy planning
• Addresses long-range dependencies in volumetric anatomy modeling
• Ensures physically consistent and structurally accurate radiation plans

Innovation

• Transformer framework for direct fluence map regression
• Two-stage design with global dose prior and beam geometry conditioning
• Physics-informed FAR loss integrating multiple fidelity and conservation constraints