基于AI的事故照射皮肤剂量快速估算模型研制

An AI-based rapid estimation model for skin dose in radiation accidents

目的基于当前核与辐射事故中皮肤剂量估算实际需求,结合神经网络,研制一个皮肤剂量快速估算模型,为放射性皮肤损伤防治提供科学依据。方法依据核与辐射事故物理剂量估算理论方法,基于最新的ICRU95号报告提供的粒子注量到皮肤吸收剂量的转换系数,构建一个基于遗传算法的BP神经网络模型,完成事故照射情况下皮肤剂量的快速估算。针对一起案例,对比蒙特卡罗模拟方法与本文所述转换系数法对皮肤剂量的估算结果,比较两者结果的一致性。结果基于遗传算法的BP神经网络模型能够完成对物理剂量估算关键转换系数的预测,整体预测误差为5.57%。在事故照射实例中,本文推荐的快速估算模型得到的皮肤吸收剂量较蒙特卡罗模拟结果相差3.39%,进一步验证了模型的可靠性和准确性。结论本文构建的皮肤剂量快速估算模型对于放射性皮肤疾病防治中所需的精准快速剂量估算具有重要意义。

Objective Based on the practical demand for skin dose estimation in current nuclear and radiation accidents,this study aims to develop a rapid skin dose estimation model using neural networks.Methods Drawing on the theoretical framework of physical dose estimation,a BP neural network model based on genetic algorithms is constructed.This model utilizes conversion coefficients from particle fluence to skin-absorbed dose,as provided in the latest ICRU Report 95.The model is designed to swiftly estimate skin doses incurred during radiation accidents.Case studies are conducted to compare the skin dose estimations obtained using Monte Carlo simulation with those derived from the proposed conversion coefficient method,assessing the consistency between the two approaches.Results The GA-BP neural network model successfully predicts conversion coefficients for physical dose estimation,with an overall prediction error of 5.57%.In the case of accidental irradiation,the skin absorbed dose obtained from the rapid estimation model recommended in this article differs by 3.39%compared to the Monte Carlo simulation results,further validating the reliability and accuracy of the model.Conclusion The rapid skin dose estimation model developed in this study is essential for for accurately and swiftly estimating doses in radiation-induced skin diseases.