Prior scanning Estimation of Dose Length Product for Chest Computed Tomography using machine learning

Abstract

Considering the radiobiological effects of X-rays, it is important to protect patients undergoing diagnostic radiological procedures by ensuring compliance with diagnostic reference levels before CT examination realization. This study investigates the use of machine learning to predict the dose-length product (DLP) in adult chest CT examinations before image acquisition. Multiple algorithms have been implemented, including Support Vector Regression, K-Nearest Neighbors, Decision Tree Regressor, and Artificial Neural Networks (ANN). The correlation map led to the elimination of the variable Age due to its low correlation with the target variable. The dataset contains 14,667 chest CT examination records for adult patients [1]. After outlier removal using a data transformation approach, the dataset was randomly divided into 80% for training and 20% for testing. The ANN model achieved the best performance, with a root mean square error (RMSE) of 33.89 mGy·cm and a coefficient of determination (R²) of 0.892. The lowest performance was observed with the Decision Tree Regressor, with an accuracy of 0.60. These findings demonstrate that machine learning models, particularly ANN and XGBoost, can provide accurate pre-scan dose estimation, potentially enabling real-time optimization of CT protocols and enhancing patient protection.