基于径向基神经网络的舒必利稳态血药浓度预测

Prediction of Steady-State Plasma Concentration of Sulpiride Based on Radial Basis Function Neural Networks

目的建立基于径向基(RBF)神经网络舒必利稳态血药浓度预测模型。方法将所收集的用于建立舒必利稳态血药浓度预测模型的数据(包括患者的性别、年龄、体重、剂量、稳态血药谷浓度、多项生理生化指标等)分为训练集、校验集和测试集,前两者用于训练RBF神经网络,后者用于测试RBF神经网络,分别利用各数据集的网络计算输出值与目标输出值之间的均方差(MSE)和相关系数(R)评价网络模型的训练效果和预测性能。结果建立以患者的性别、年龄、体重、剂量、多项生理生化指标等37项参数为输入变量,舒必利稳态血药浓度为输出变量的37-1-1结构的RBF神经网络,当网络中心宽度SP值为2.3时,训练集、校验集和测试集的MSE分别为4.50×10-6、0.003 531和0.011 001,R值分别为0.999 91、0.955 32和0.814 25。结论利用RBF神经网络所建立的舒必利稳态血药浓度预测模型的预测效果较好,但泛化能力尚待提高。

Objective To establish a model for predicting the steady-state plasma concentration of sulpiride based on radial basis function （RBF） neural network. Methods The data （ including the patients＇ gender, age, dose, weight, dosage, steady-state plasma trough concentration and multiple physiological and biochemical indexes, etc. ） to modeling the steady-state plasma concentration of sulpiride was divided into training set, validation set and test set. The first 2 was used for training the RBF neural network, the latter was used for testing the RBF neural network. The error of mean square （MSE） and coefficient correlation （R） between the computed output value and objective output value of every data sets was used for evaluating the training and predictive effect of the RBF neural network, respectively. Results The 37-1-1 RBF neural network model was established which had captured the relationships between the input variables （the patients＇ gender, age, dose,weight, dosage and multiple physiological and biochemical indexes etc. 37 parametes） and the output variable （ the steady-state plasma concentration of sulpiride）. When SP value, that is the network center width,was 2.3,the MSE and R values of the training set,validation set and test set were 4.50 x 10 -6,0. 003 531,0.011 001 ,and 0. 999 91,0. 955 32,0. 814 25, respectively. Conclusion The RBF neural network model has the better predictive effect to predict the steady-state plasma concentration of sulpiride. But it＇ s generalization is required to be improved.