Author: Parker Anderson, Elizabeth L. Bossart, Jonathan Cyriac, Nesrin Dogan, Robert Kaderka, Yihang Xu 👨🔬
Affiliation: University of Miami, University of Miami, Sylvester Comprehensive Cancer Center, University of Miami Sylvester Comprehensive Cancer Center 🌍
Purpose:
Knowledge-based planning (KBP) can enhance the treatment planning process in cancer radiotherapy (RT). By training a KBP model with high-quality treatment plans developed by experts, dose-volume histogram (DVH) and optimization objective estimations can be efficiently generated to assist in the inverse planning process for intensity-modulated proton therapy (IMPT). In this study, we developed and fine-tuned a KBP-based model to optimize IMPT treatment for breast cancer using a novel iterative tuning process.
Methods:
Fifty patients previously treated with IMPT to the breast/chest wall were used to train our KBP model (25 left breast and 25 right breast). A subset of ten patients (5 left-sided and 5 right-sided) underwent an iterative tuning process to validate the integrity of the plans generated by our model. The plans were created and optimized solely using the objective list generated by the model. The quality of these tuned plans was compared to manually optimized clinical plans using DVH-based metrics for target volumes and organs at risk (OARs).
Results:
Most of the DVH metrics investigated showed no statistically significant differences between the clinical and model-generated plans. The only variation observed in these ten tuning cases was slightly better sparing of the Heart at Dmean (-0.18 ± 0.23 Gy) in the model-generated plans. However, all metrics for the model-generated plans remained within clinically acceptable limits.
Conclusion:
The breast KBP model developed in this study shows promise for future clinical applications. Most plans generated with this model were clinically acceptable without further adjustments. Our novel iterative tuning process enhanced the outcomes of plans produced by the model. While the model still requires validation with a separate patient cohort outside of the training set—results of which will be reported in the future—it demonstrates the potential of a KBP model for creating IMPT plans in breast cancer treatment.