Feasibility study of predicting dose of radioiodine in hyperthyroidism patients based on neural network / 中华放射医学与防护杂志

ABSTRACT

Objective:To construct back propagation (BP) neural network model to predict the dose required for 131I therapy for hyperthyroidism and to calculate the personalized dose plan for patients. Methods:A complete set of data of patients treated for hyperthyroidism radioaiodine was collected from the nuclear medicine departments of several medical colleges in Shanghai, including history, examination result, treatment course, etc. As a result, a prediction model was established. The predicated result for BP neural network, radial basis function (RBF) neural network and Support Vector Machine (SVM) were compared by means of small sample data. The optimal model was selected to predict administrated dose and to finally test the accuracy of the model.Results:The average errors in BP neural network, RBF neural network and SVM model based on small samples were 5.53%, 7.09% and 9.64%, respectively. After comparison, BP neural network was selected to build the prediction model. 30 cases of data were selected by random sampling to verify the BP neural network. The mean error, mean square error, minimum error and maximum error of the prediction result were 7.22%, 0.053, 0.57% and 13.78%, respectively.Conclusions:In this study, a neural network prediction method was proposed to provide a more accurate dose for patients in need of radioiodine therap for hyperthyroidism, and to reduce the possibility of radiation damage or the unsatisfactory therapeutic effect caused by insufficient dose. It has clinical practical significance in providing the reference for clinicians to evaluate the administrated dose.