Jointly trained machine learning models for automatic contouring in radiotherapy applications
Inventors
Novosad, Philip P. • Beriault, Silvain
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12728284-B2
Publication Date
2026-09-08
Expiration Date
Abstract
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.
Core Innovation
The invention jointly trains first and second machine-learning models using images from independent training datasets. The models share at least one learned model parameter while retaining parameters specific to each model. When trained, the models automatically contour images for radiotherapy treatment, accommodating datasets that differ in anatomy, imaging contrast, scan time, surgery status, and generated structures.
For treatment planning, a model is selected based on an image parameter and applied to a subject image. The selected model contours anatomical structures without user intervention using both shared and model-specific learned parameters. The resulting contours are subsequently processed to generate a radiotherapy treatment plan.
Claims Coverage
The claims cover three independent claims directed to two computer-implemented methods and one radiotherapy system, comprising three inventive features.
Joint training of machine-learning models
The first independent claim covers jointly training first and second machine-learning models using images from independent training datasets. The models share at least one learned model parameter and, when trained, automatically contour images for radiotherapy treatment.
Model selection, contouring, and treatment-plan generation
The second independent claim covers obtaining a subject image, selecting a previously trained model based on an image parameter, and applying the model using shared and model-specific learned parameters. The model contours anatomical structures without user intervention and the contours are processed to generate a radiotherapy treatment plan.
Radiotherapy system using a jointly trained model
The third independent claim covers a radiotherapy system comprising a radiation therapy device and a processor. The processor selects and applies a previously trained model based on an image parameter, contours anatomical structures without user intervention, and processes the contours to generate a radiotherapy treatment plan.
Together, the independent claims cover joint training of models across independent datasets, image-based model selection and automatic contouring for treatment-plan generation, and a radiotherapy system configured to execute the plan-generation process.
Stated Advantages
Joint training leverages commonalities across datasets to improve model performance while requiring less training data per dataset, as demonstrated by an improved bladder contour in a female-pelvis comparison after joint training with additional male-pelvis data.
Documented Applications
Automatic contouring of anatomical structures in subject images for generating radiotherapy treatment plans.
Interested in licensing this patent?