Prediction of Breast Radiation Absorbed Dose Chest CT Examinations Using Machine Learning Techniques

Area of Science:

• Medical Imaging
• Radiology
• Machine Learning in Healthcare

Background:

• The breast is sensitive to radiation during chest CT scans, increasing cancer risk.
• Accurate, patient-specific dose estimation is crucial for radiation safety.
• This study explores machine learning for personalized breast radiation dose prediction.

Purpose of the Study:

• To estimate breast radiation dose during chest CT using a machine learning (ML) personalized prediction approach.
• To develop and evaluate ML models for predicting patient-specific breast radiation doses.

Main Methods:

• Retrospective analysis of 653 female patients undergoing mammography and chest CT.
• Utilized demographic and anatomical data (BMI, breast thickness) and dose length product (DLP).
• Implemented and compared five ML algorithms: CatBoost, Gradient Boosting, Extra Trees, AdaBoost, and Random Forest.

Main Results:

• The CatBoost algorithm optimized with Particle Swarm Optimization (CatBoostPSO) showed the best performance.
• CatBoostPSO achieved the lowest MSE (0.3795), MAE (0.3846), and MAPE (4.37%), with the highest R² (0.9875).
• Ensemble and optimized models effectively predicted breast radiation dose.

Conclusions:

• A machine learning framework provides rapid, accurate breast radiation dose estimation for chest CT.
• This patient-specific method supports personalized radiation dose monitoring and optimization.
• The approach enhances radiation safety in clinical practice.