Prediction of dose to the relatives of patients treated with radioiodine-131 using neural networks

Abstract

In this study, the effective dose received by the family members and caregivers of 52 thyroid cancer patients, who had been treated with radioiodine I-131, was measured to investigate the ability of the neural network to predict the doses to the relatives. The effectiveness of this method to predict the relatives who will receive doses of more than 1 mSv was evaluated. The effective doses were measured by TLD. The inputs of the neural network include 13 different parameters that can potentially affect the dose, and the output was the dose to the family members. The neural networks in this study were feed-forward with a sigmoid activation function and one hidden layer. The mean and median of the measured doses were 0.45 and 0.28 mSv and its range was 0.1-3.64 mSv. The mean square error of the predicted doses by the neural network and the measured doses by TLD (mean squared error) for 99 individuals was 0.142. The optimum neural network was able to predict all the relatives who received doses of more than 1 mSv. The area under the receiver operating characteristic curve for the trained neural network was 0.957, showing its ability to distinguish these groups. Predicting the dose to a patient's relatives before release is a helpful strategy for future optimisation. Using neural networks is a promising method for predicting the dose to the family members and defining high-risk patients and relatives. Patient-specific criteria for release and patient-specific advice and consultation can be used to reduce the dose to each family member.