Joint Deep Neural Network Training for Radiotherapy OAR Contouring

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The challenge in radiotherapy treatment planning is the time-consuming and variable nature of manual contouring of organs-at-risk (OARs), which is compounded by the need for large training datasets for accurate machine learning models, leading to high costs and inefficiencies.

Innovation Solution

Joint training of multiple machine learning models across clinical datasets to leverage commonalities and reduce the required training data per dataset, enabling more performant automatic contouring.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual contouring is used to delineate target tumors and organs-at-risk, then accuracy can be maintained, but the process becomes extremely time-consuming and labor-intensive

Engineering Contradiction:
Improvecontouring accuracyVSAvoidtreatment planning time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent replaces the manual mechanical contouring process with an automated machine learning system. Deep neural networks are trained on labeled training data to automatically delineate target tumors and organs-at-risk on medical images, eliminating the need for manual tracing while maintaining clinical accuracy. The system uses automated algorithms to identify and segment anatomical structures, transforming a labor-intensive manual task into an efficient computational process.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Reliability

If separate machine learning models are trained for each clinical dataset, then model specificity is improved, but the amount of training data required increases significantly

Engineering Contradiction:
Improvemodel accuracyVSAvoidtraining data volume
Core Design Contradiction:
ReliabilityVSQuantity of substance

Solution Approach 1:

The patent combines multiple clinical datasets into a unified training framework. Instead of training separate models for each dataset, the system merges data from multiple sources to create a comprehensive training set. This approach allows the model to learn from diverse anatomical variations and imaging protocols while reducing the per-dataset data requirement. The merged training data enables the model to generalize across different clinical scenarios with smaller individual dataset contributions.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent creates a universal machine learning model that can handle multiple clinical datasets and imaging types. The trained model is designed to be multi-functional, capable of delineating various anatomical structures across different datasets and imaging modalities. This universal model eliminates the need for separate specialized models for each dataset, reducing overall training data requirements while maintaining adaptability to different clinical scenarios through joint training on diverse data.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Adaptability or versatility

If multiple separate machine learning models are trained for different datasets, then model versatility is improved, but the overall system complexity increases

Engineering Contradiction:
Improvedataset compatibilityVSAvoidsystem complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent implements a universal model architecture that can process multiple clinical datasets through a single unified system. The model is designed with multi-functionality, capable of handling different imaging modalities and anatomical variations without requiring separate specialized models. This universal approach maintains versatility across datasets while significantly reducing system complexity compared to managing multiple separate models.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent segments the training process into distinct phases while maintaining a unified model architecture. The system divides training data into different datasets for organized processing, but integrates them through a single model framework. This segmentation of the training approach allows the system to handle diverse datasets systematically while avoiding the complexity of maintaining multiple separate models, as each dataset contributes to training the same universal model.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4279125B1Joint training of deep neural networks across clinical datasets for automatic contouring in radiotherapy applications
Publication Date: 2025.11.05 ELEKTA AB
  • EP4279125B1 patent drawingFigure 1
  • EP4279125B1 patent drawingFigure 2A
  • EP4279125B1 patent drawingFigure 2B

AI summary

Joint training techniques to train multiple models across clinical datasets for automatic contouring. Rather than using separate deep neural networks that are trained independently for each different dataset (e.g., a different image contrast or anatomy), joint training can be used to train multiple models simultaneously across clinical datasets for automatic contouring. By taking advantage of commonalities between two or more datasets, the techniques effectively take advantage of data that would otherwise be considered irrelevant to the task - allowing the user to train more performant models while requiring less training data per dataset.