Development of a Machine Learning Model for Optimal Applicator Selection in High-Dose-Rate Cervical Brachytherapy
Kailyn Stenhouse1,2, Michael Roumeliotis1,2,3, Philip Ciunkiewicz4
1Department of Physics and Astronomy, University of Calgary, Calgary, AB, Canada.
Frontiers in Oncology
|March 22, 2021
Summary
A machine learning model was developed to aid in selecting applicators for high-dose-rate cervical brachytherapy. The model accurately predicts the best applicator, improving treatment planning.
Area of Science:
- Oncology
- Medical Physics
- Machine Learning
Background:
- High-dose-rate (HDR) cervical brachytherapy requires careful selection between intracavitary (IC) and hybrid interstitial (IS) applicators.
- Accurate applicator selection is crucial for optimizing dose delivery and minimizing toxicity.
Purpose of the Study:
- To develop and validate a preliminary machine learning (ML) model for selecting between IC and IS applicators in HDR cervical brachytherapy.
- To identify key geometric features that predict optimal applicator choice.
Main Methods:
- A dataset of 233 HDR cervical brachytherapy treatments was analyzed.
- Geometric features of the HR-CTV and organs at risk (OARs) were extracted.
- Feature selection and 12 classification algorithms were employed, with the top three combined using soft voting for the final model.
Main Results:
- HR-CTV volume and mean lateral extent were the most important features for applicator selection.
- The final voting model achieved a high predictive accuracy of 91.5 ± 0.9% and an F1 Score of 90.6 ± 1.1%.
- Tree-based ensemble methods, including AdaBoost, Gradient Boosting, and Random Forest classifiers, performed best individually.
Conclusions:
- The developed ML model demonstrates strong discriminative performance in applicator selection.
- This model holds potential for prospective clinical use in HDR cervical brachytherapy treatment planning.
- Further clinical validation is recommended to confirm the model's utility.
