PREDICTIVE MODELING OF PATIENT RADIATION DOSE IN CONVENTIONAL RADIOGRAPHY IN NIGERIAN HOSPITALS

Abstract

This study evaluated the patient’s radiation doses in conventional radiography at three government hospitals in Delta State, Nigeria and established predictive models for estimating the entrance skin dose (ESD) from routine exposure parameters. Data of 761 adult patients examined in abdomen, chest, lumbosacral region, pelvis and skull were analyzed. The ESD was measured with a calibrated Unfors Multi-O-Meter 710L with a water phantom according to the IAEA (2025). The mean ESD value was observed to be the highest in lumbosacral examinations (3.73±1.08 mGy) and the lowest in chest examinations (0.42±0.38 mGy). Pearson correlation analysis showed a very strong positive correlation between ESD and kerma-area product (r=0.963) and between ESD and the kVp×mAs product (r=0.945). Linear regression modeling exhibited excellent predictive performance, with coefficient of determination and Root Mean Square Error of (R²=0.937 and RMSE=0.425 mGy). Logistic regression classification accurately classified low- versus high-dose examinations (median threshold 1.60 mGy) with 96.7% accuracy. The local DRLs based on the 75th percentile of the ESD distributions compared well to international benchmarks. Results demonstrate that patient ESD can be predicted from routinely recorded exposure parameters with good reliability, using simple interpretable machine learning models, providing a low-cost and scalable decision-support tool for dose optimization in resource-limited healthcare settings.