Machine-Learned Objective Functions for Radiation Treatment Planning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing treatment planning systems for radiation therapy struggle with ill-posed problem formulations, making it difficult for users to translate desirable dosimetric characteristics and treatment goals into mathematical objective functions.

Innovation Solution

A machine learning-based system that trains a model to convert clinical objectives into computable objective functions using user-ranked previous treatment plans, allowing users to customize and generate precise mathematical functions for radiation therapy planning.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If traditional mathematical objective functions are used in treatment planning systems, then the optimization process can be performed, but users cannot directly translate intuitive clinical goals into appropriate mathematical formulations

Engineering Contradiction:
Improveease of translating clinical goalsVSAvoidprecision of objective function formulation
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent introduces machine learning models as an intermediary between clinical goals and mathematical objective functions. The system uses trained ML models to automatically translate intuitive clinical objectives into computable mathematical formulations, eliminating the need for users to manually formulate complex objective functions while maintaining precision through data-driven learning from expert treatment plans

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces the manual mechanical process of mathematical formulation with an automated intelligent system. Instead of users manually translating clinical goals into mathematics, the system uses machine learning algorithms to perform this translation automatically, substituting human cognitive effort with computational intelligence that has been trained on expert data

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

2Manufacturing precision

If treatment planning systems provide detailed objective functions and optimization capabilities, then plan quality can be improved, but the complexity of the system increases making it harder to use

Engineering Contradiction:
Improveprecision of treatment planVSAvoidcomplexity of planning system
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent enables the system to serve itself by automatically generating objective functions from clinical goals without requiring user expertise in mathematical formulation. The machine learning models are self-trained on expert data and can independently translate new clinical objectives into appropriate mathematical formulations, making the complex system easier to use while maintaining high precision

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent performs preliminary training of machine learning models using expert treatment plans before actual treatment planning. This preliminary action pre-computes the translation between clinical goals and objective functions, so that during actual use, the system can directly apply these learned translations without requiring users to understand the underlying mathematical complexity

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4433157B1A system for generating objective functions for treatment planning
Publication Date: 2025.10.01 ELEKTA AB
  • EP4433157B1 patent drawingFigure 1
  • EP4433157B1 patent drawingFigure 2
  • EP4433157B1 patent drawingFigure 3

AI summary

System (OGS) and related method for generating an objective function for use in radiation treatment, RT, planning. The system may include a machine learning model (M) and a training system (TS) for training the model (M) based on training data. The training data may include previous (at least partial) RT plans, and a user awarded ranking thereof. The model, once trained, may be used as the objective function. The system allows a user to turn, in a defined manner, clinical objectives or goals into a computable objective function which can be used for RT planning.