Related Experiment Video
Updated: May 27, 2026
08:25
Radiation Planning Assistant - A Streamlined, Fully Automated Radiotherapy Treatment Planning System
Published on: April 11, 2018
A hybrid cluster-then-predict machine learning radiotherapy knowledge-based planning framework for similarity
Trent Benedick1, J Daniel Chu2, Stephanie Zhou2
1Image Processing and Informatics Laboratory, Department of Biomedical Engineering, University of Southern California, 1042 Downey Way, Los Angeles, CA, 90089, USA. benedick@usc.edu.
Radiation Oncology (London, England)
|May 26, 2026
Summary
This study introduces a new knowledge-based planning (KBP) algorithm that uses geometric case matching to identify relevant radiotherapy planning templates. The novel approach improves plan quality and standardization for clinicians.
Area of Science:
- Radiation Oncology
- Medical Physics
- Machine Learning in Healthcare
Background:
- Current radiotherapy treatment planning relies on individual experience and existing knowledge-based planning (KBP) models.
- There is a need for improved methods to assist clinicians in creating high-quality, standardized treatment plans.
Purpose of the Study:
- To introduce and validate a novel KBP algorithm for radiotherapy treatment planning.
- To match retrospective cases based on holistic target-to-Organ at Risk (OAR) geometry.
- To provide planning templates by identifying similar past cases.
Main Methods:
- Developed a multi-step algorithm to calculate geometric similarity between radiotherapy cases.
- Extracted quantitative geometric features (Overlap Volume Histograms, Spatial Target Signatures) and calculated dissimilarities using Earth Mover's Distance (EMD).
- Employed a hybrid machine learning approach (Mixture-of-Experts) with spectral clustering to predict case similarity.
Main Results:
- Validated the algorithm with 192 Head and Neck Cancer cases.
- Demonstrated a strong correlation (r=0.8009) between the KBP model's geometric similarity score and actual dose similarity.
- Achieved a high Area Under the Curve (AUC) of 0.8797 in clinical evaluation, confirming the model's ability to identify relevant reference cases.
Conclusions:
- Validated a novel KBP algorithm for identifying clinically relevant retrospective cases using target-OAR geometric constellations.
- The algorithm provides personalized, knowledge-based reference data to assist clinicians.
- This tool enhances standardization, improves plan quality, and aids initial treatment plan creation.
