Dosimetric parameter-based machine learning to predict radiation pneumonitis following breast radiotherapy

Area of Science:

• Radiation oncology
• Medical physics
• Machine learning in healthcare

Background:

• Radiation pneumonitis (RP) is a significant toxicity after breast radiotherapy.
• Predicting which patients will develop symptomatic RP remains challenging despite modern techniques.
• Accurate prediction is crucial for personalized treatment planning and patient management.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting clinically significant RP.
• To utilize ipsilateral-lung dosimetric parameters for prediction.
• To assess the role of high-dose versus low-dose lung exposure in RP risk.

Main Methods:

• Retrospective study of 184 breast cancer patients treated with 3D-CRT.
• Extraction of ipsilateral lung dose-volume metrics, Dmax, MLD, and Deff.
• Training and cross-validation of four ML models (LightGBM, random forest, SVM, logistic regression).

Main Results:

• 28% of patients developed clinically significant RP (CTCAE grade >=2).
• All four ML models demonstrated strong predictive performance (AUC >0.8).
• High-dose lung parameters (e.g., V45Gy, Dmax) were the most significant predictors.

Conclusions:

• ML models using routine lung dosimetric data can predict RP post-breast radiotherapy.
• Higher doses to smaller lung regions are more critical for RP risk than low-dose exposure.
• This approach can aid personalized treatment planning and identify at-risk patients for closer monitoring.