Comparison and Optimization of Machine Learning Regression Models for Predicting Electromagnetic Radiation Exposure in Patients Undergoing Coronary Angiography

This study aims to propose an optimal regression model for predicting electromagnetic radiation exposure in patients undergoing coronary angiography (CAG). A retrospective analysis of data from 116 CAG patients was conducted, using age, gender, BMI, puncture site, and equipment type as independent variables, with dose area product (DAP) as the dependent variable. Three regression models - Random Forest, XGBoost, and Linear - were implemented, and optimal hyperparameters were explored. The Random Forest model demonstrated the best predictive performance, with predicted DAP values converted to effective doses for clinical utility (RMSE: 1.1353 mSv, MAE: 0.9 mSv, R² Score: 0.5766). Residual analysis confirmed a normal distribution (Kolmogorov- Smirnov test, p > 0.05). The developed model can enhance the informed consent process, ensure patients' right to information, and aid in establishing optimal examination protocols by providing anticipated electromagnetic radiation doses for individual patients prior to CAG procedures.