Breast radiotherapy planning: A decision-making framework using deep learning.

Area of Science:

• Medical Physics
• Radiotherapy
• Artificial Intelligence in Medicine

Background:

• Optimizing breast cancer radiotherapy involves balancing tumor eradication with minimizing radiation dose to healthy tissues.
• Selecting between intensity-modulated radiation therapy (IMRT) and three-dimensional conformal radiation therapy (3D-CRT) is critical and depends on patient-specific anatomy and dosimetric constraints.

Purpose of the Study:

• To develop a deep learning-based decision-making framework for predicting radiation dose distributions.
• To aid in the selection of optimal radiotherapy techniques (IMRT vs. 3D-CRT) for breast cancer patients.

Main Methods:

• A 2D U-Net convolutional neural network (CNN) was employed to predict dose distribution maps and dose-volume histogram (DVH) metrics.
• The model was trained on retrospective data from 346 patients across two institutions, addressing variations in CT systems and clinical practices.
• External validation was performed on 30 patients, with plans evaluated by an independent medical physicist using confusion matrices to compare model predictions with clinical decisions.

Main Results:

• The deep learning model achieved high concordance in predicting dose distributions and DVH metrics for both IMRT and 3D-CRT, particularly for organs at risk (OARs).
• The decision-making framework demonstrated superior performance compared to historical methods, achieving 90% accuracy, 95.7% recall, and 91.7% precision against independent clinical evaluations.
• In contrast, historical decision-making showed lower accuracy (50%), recall (47.8%), and precision (78.6%).

Conclusions:

• The developed deep learning model accurately predicts dose distributions for both 3D-CRT and IMRT breast cancer treatments.
• The framework ensures reliable estimation of OAR doses and significantly enhances decision-making accuracy, recall, and precision compared to traditional approaches.