Machine Learning Dose Distribution Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current radiotherapy dose distribution simulations, such as those using pencil beam and cone convolution algorithms, lack accuracy, while Monte Carlo algorithms provide high accuracy but at the cost of computation speed, necessitating a method for generating accurate and efficient dose distributions.
Innovation Solution
A system and method utilizing a trained machine learning model to improve the accuracy of initial dose distributions, where personalized data from a subject, including radiotherapy treatment plans and density distributions, is used to generate a second dose distribution with higher accuracy and faster computational speed by iteratively updating machine learning model parameters based on training samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Monte Carlo algorithm is used to calculate radiation dose distribution, then accuracy is improved, but computation speed deteriorates
Solution Approach 1:
The patent pre-calculates and stores dose distribution data for various radiation conditions using Monte Carlo algorithm during training. When actual treatment planning is needed, the system queries pre-computed results from the database, avoiding real-time complex calculations and thus achieving fast response with high accuracy.
Solution Approach 2:
The patent creates a simplified representation (digital twin) of the patient's anatomy and treatment scenario using machine learning models trained on Monte Carlo simulation data. This virtual copy allows rapid dose calculation without performing computationally intensive physical simulations for each treatment plan.
2Productivity
If pencil beam algorithm or cone convolution algorithm is used to calculate radiation dose distribution, then computation speed is improved, but accuracy deteriorates
Solution Approach 1:
The patent introduces machine learning models as intermediaries between simple fast algorithms and complex accurate Monte Carlo simulations. The ML models learn the relationship between simplified input parameters and accurate dose distributions from training data, enabling fast predictions that capture essential physical effects without performing full Monte Carlo calculations.
Solution Approach 2:
The patent transforms the input parameters for dose calculation from detailed physical simulation parameters to simplified clinical parameters (e.g., beam energy, field size, patient anatomy). The machine learning model handles the complex parameter transformation and dose calculation, achieving both speed and accuracy by operating in the simplified parameter space.
Data Source
AI summary
A system for generating a dose distribution is provided. The system may obtain a first dose distribution in at least a portion of a subject. The system may also obtain a trained machine learning model. The system may further generate, based on the first dose distribution and the trained machine learning model, a second dose distribution in the at least a portion of the subject, wherein the second dose distribution has a higher accuracy than that of the first dose distribution.