# Knowledge-based planning for intensity-modulated proton therapy of the breast and chest wall with regional lymph nodes: development, validation, and comparative evaluation of two models

## TL;DR

This study develops and compares two knowledge-based planning models for proton therapy in breast cancer treatment, showing they can generate acceptable treatment plans efficiently.

## Contribution

The study introduces and validates two KBP models for IMPT in breast cancer with regional lymph nodes, demonstrating their clinical feasibility and efficiency.

## Key findings

- Both KBP models generated plans meeting clinical constraints with comparable target coverage to clinical plans.
- The side-specific model showed better sparing of organs like the esophagus and thyroid compared to the combined model.
- Physicians preferred KBP plans in 11 out of 20 cases, indicating clinical acceptability.

## Abstract

Knowledge-based planning (KBP) can improve efficiency and consistency in radiotherapy. In intensity-modulated proton therapy (IMPT) for breast cancer, particularly with regional lymph nodes, treatment planning remains complex. This study developed, validated, and compared two KBP models for IMPT of the breast and chest wall (CW) to enhance plan quality, reduce inter-planner variability, and streamline workflows.

Fifty patients (25 left-sided, 25 right-sided) previously treated with IMPT to the breast or CW including regional lymph nodes were used for model development. A combined model was trained on all patients, while side-specific models were developed separately for left- and right-sided treatments. Planning objectives were refined via iterative replanning of 20 training cases. For validation, 20 additional patients (10 left, 10 right) were evaluated using the respective models. Dosimetric metrics for target volumes and organs at risk (OARs) were compared among model-generated and clinical plans using paired t-tests (p < 0.05). An expert physician ranked each plan for clinical preference.

All KBP-generated plans met clinical constraints. Both models achieved target coverage comparable to clinical plans. The combined model produced cooler hot spots in axillary, supraclavicular, and internal mammary nodes, with modest improvements in OAR sparing (heart, esophagus, trachea, thyroid). The side-specific model showed similar benefits, particularly for esophagus and thyroid sparing. Between models, side-specific plans had cooler hot spots, while the combined model offered slightly better OAR sparing. Physician review deemed all KBP plans clinically acceptable, preferring the KBP plan in 11 of 20 cases.

KBP models can generate clinically acceptable IMPT plans for breast and CW with regional lymph nodes, achieving comparable target coverage to clinical plans with modest improvements in OAR sparing and hot spot reduction, supporting their potential to streamline treatment planning.