Diffusion Fluence Map Generation for Accurate Radiation Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating fluence maps in radiation therapy lack the ability to accurately predict detailed information from CT images, resulting in fluence maps that are too smooth and differ significantly from real images, necessitating skilled planners and extensive optimization time for effective dose distribution.
Innovation Solution
Employing a denoising diffusion probabilistic model as a deep learning model to generate high-quality fluence maps by inputting radiation information at each beam angle into a fluence map generative model, utilizing likelihood estimation to improve the accuracy and quality of fluence map generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a convolutional neural network is used to predict fluence map from CT image, then the generation speed is improved, but the prediction accuracy deteriorates resulting in fluence maps that are too smooth and differ significantly from real images
Solution Approach 1:
The patent introduces an intermediary refinement process between the initial CNN prediction and the final fluence map. The method uses the predicted fluence map as an initial guess and iteratively refines it by comparing with actual dose distribution and adjusting through optimization algorithms. This intermediary step bridges the gap between fast but inaccurate CNN prediction and accurate but slow traditional methods.
Solution Approach 2:
The patent performs preliminary action by using the convolutional neural network to generate an initial fluence map prediction before the final optimization. This preliminary fluence map serves as a head start for subsequent refinement processes, reducing the computational burden and iteration time required to achieve accurate results compared to starting from scratch with traditional optimization methods.
2Manufacturing precision
If traditional optimization methods are used to generate accurate fluence maps, then the fluence map quality is improved, but the optimization time increases significantly
Solution Approach 1:
The patent applies preliminary action by using the convolutional neural network to generate an initial fluence map prediction that is already close to the optimal solution. This preliminary prediction significantly reduces the number of iterations required by subsequent optimization algorithms, thereby maintaining high fluence map quality while dramatically reducing the total optimization time compared to traditional methods that start from scratch.
Solution Approach 2:
The patent implements local quality by applying different processing strategies to different aspects of fluence map generation. The CNN model handles the global structure and general pattern prediction efficiently, while localized refinement algorithms focus computational resources on specific regions requiring higher precision. This division allows the system to achieve high overall quality without uniformly applying computationally expensive optimization across the entire fluence map.
3Reliability
If skilled planners are involved to adjust fluence maps for effective dose distribution, then the treatment plan quality is improved, but the operational complexity increases
Solution Approach 1:
The patent implements self-service by enabling the system to automatically generate and optimize fluence maps with minimal human intervention. The integrated approach combines AI prediction with automated optimization algorithms that can independently adjust fluence maps to achieve effective dose distribution. This reduces reliance on skilled planners while maintaining treatment plan quality, thereby reducing operational complexity.
Solution Approach 2:
The patent incorporates feedback mechanisms where the system continuously evaluates the generated fluence maps against desired dose distribution criteria and automatically adjusts parameters to improve results. This closed-loop feedback system ensures high treatment plan quality by iteratively refining the fluence maps based on performance metrics, reducing the need for manual expert intervention and simplifying the overall planning process.
Data Source
AI summary
A fluence map generation method includes: determining radiation information of a target volume at each beam angle of a plurality of beam angles, the target volume including a target area and organs at risk; and inputting the radiation information of the target volume at each beam angle into a fluence map generative model based on a denoising diffusion probabilistic model to obtain a fluence map corresponding to each beam angle. The fluence map corresponding to each beam angle is used to indicate an intensity of each beamlet corresponding to each beam angle in the target volume.