Anatomical Parameter-driven Volumetric Modulated Arc Therapy Optimization in Left-sided Breast Cancer: A Machine...

Area of Science:

• Radiation Oncology
• Medical Physics
• Machine Learning in Healthcare

Background:

• Radiotherapy for left-sided breast cancer requires careful lung dose management.
• Volumetric modulated arc therapy (VMAT) offers advanced dose delivery options.
• Predictive modeling can improve treatment planning accuracy.

Purpose of the Study:

• To compare the efficacy of different VMAT techniques (VMAT-2P, VMAT-4P, VMAT-5P) for left-sided breast cancer.
• To develop machine learning-based predictive models for lung doses.
• To identify key anatomical predictors of lung dose in this patient cohort.

Main Methods:

• Retrospective analysis of 101 left-sided breast cancer patients.
• Comparison of three VMAT techniques: VMAT-2P, VMAT-4P, and VMAT-5P.
• Development of lung dose predictive models using regression algorithms (LASSO, Ridge, Linear Regression, Random Forest Regressor).

Main Results:

• VMAT-4P demonstrated the lowest mean lung dose and V20 Gy values.
• VMAT-5P showed the lowest exposure to low lung doses (V5 Gy).
• Central lung distance (CLD) was the strongest predictor of lung dose; Random Forest Regressor yielded the best predictive performance for mean lung dose.

Conclusions:

• VMAT-4P is recommended for left-sided breast cancer radiotherapy due to its balance of target coverage and organ-at-risk sparing.
• Machine learning, particularly the Random Forest Regressor, effectively predicts mean lung dose, enhancing treatment planning.