Augmented Radiotherapy Generative Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep learning methods for radiotherapy dose calculations in photon radiation therapy are too time-consuming for clinical adoption and often solve a proxy problem, lacking assurance in clinical validation of dose accuracy and quality.
Innovation Solution
A device that trains an augmented radiotherapy generative model using a Gamma index Passing Rate (GPR) loss function, enabling the generation of accurate and high-quality dose distributions from input dose distributions, optimizing clinically relevant properties and preventing gradient vanishing issues through a differentiable function and annealing schedule.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Monte Carlo simulation is used for radiotherapy dose calculations, then dose calculation accuracy is improved, but computation time increases significantly
Solution Approach 1:
The patent creates a neural network model that learns to replicate the results of computationally intensive Monte Carlo simulations. The network is trained on input-output pairs from Monte Carlo simulations, allowing it to generate dose distributions quickly without performing the full simulation each time. This copying approach enables fast inference while maintaining accuracy comparable to the gold standard Monte Carlo method.
Solution Approach 2:
The patent performs preliminary training of the neural network model using Monte Carlo simulation data during the development phase. Once trained, the model can rapidly generate dose distributions for new treatment plans without repeating the time-consuming Monte Carlo simulations. This preliminary action of training allows for fast subsequent applications in clinical settings.
2Loss of time
If common computer vision loss functions are used for training deep learning models, then training time is reduced, but clinical validation assurance is compromised
Solution Approach 1:
The patent modifies the loss function parameters and structure to incorporate clinically relevant metrics such as gamma index passing rate, dose coverage, and dose homogeneity. Instead of using generic computer vision loss functions, the patent implements a customized loss function that directly optimizes for clinical validation criteria, ensuring that the trained model produces dose distributions that meet clinical standards.
Solution Approach 2:
The patent implements a feedback mechanism where the loss function continuously evaluates the generated dose distributions against clinical validation criteria. The gamma index passing rate and other clinical metrics provide feedback during training, allowing the model to iteratively improve its output to meet clinical requirements. This feedback loop ensures that the model learns to produce clinically valid results.
3Device complexity
If proxy loss functions are used for training, then training process is simplified, but dose distribution quality cannot be strictly optimized
Solution Approach 1:
The patent changes the loss function parameters to directly reflect clinical validation requirements. The loss function incorporates parameters such as gamma index passing rate thresholds, dose coverage percentages, and dose homogeneity metrics. This parameter transformation allows the training process to directly optimize for dose distribution quality rather than using abstract proxy objectives.
Solution Approach 2:
The patent segments the loss function into multiple components, each corresponding to specific clinical validation criteria (e.g., dose coverage loss, dose homogeneity loss, gamma index loss). This segmentation allows for targeted optimization of different aspects of dose distribution quality while keeping the overall training process manageable through modular loss function design.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates to a device for obtaining a trained augmented radiotherapy generative model, configured to generate augmented radiotherapy dose distributions from input, previously calculated, radiotherapy dose distributions, said device comprising: - at least one input configured to receive a training data set comprising at least one training sample, - at least one processor configured to define said augmented radiotherapy generative model by training an untrained generative model using said training dataset and a GPR loss function, - at least one output configured to provide as output said trained augmented radiotherapy generative model.