目的:评价用径向基(RBF)神经网络所建立的预测氯氮平稳态血药浓度模型的预测性能.方法:将数据分为训练集、校验集和测试集来建立获取输入、输出变量两者间关系的RBF网络模型,其中以患者的性别、年龄、体重、剂量、血压、多项生理生化指标等37项参数为输入变量,氯氮平稳态血药浓度为输出变量.用训练集和校验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)来综合评价网络模型的学习效果,用测试集的网络计算输出值与目标输出值之间的MSE和R来评价网络模型的预测性能.结果:当扩展系数(SP)值为3.0时,训练集的MSE为1.33 ×10(-5),R值为0.99985,校验集的MSE为0.002 833,R值为0.971 86,测试集的MSE为0.005 439,R值为0.93676,网络模型的预测效果和泛化能力较好.结论:RBF网络用于预测氯氮平稳态血药浓度的研究是可行和有效的.%OBJECTIVE: To evaluate the performance of a model for predicting the steady-state plasma concentration of clozapine established by radial basis function (RBF) neural network. METHODS: The data was divided into training set, validation set and test set to establish the RBF neural network model which had obtained the relationships between input variables and output variable. Input variables included 37 parameters, such as patients' gender, age, body weight, dosage, blood pressure and multiple physiological and biochemical indexes. Output variable was steady-state plasma concentration of clozapine. The effect of RBF neural network model was evaluated comprehensively using mean square (MSE) and coefficient correlation (R) between the computed output value and objective output value of training set and validation set. And predictive performance of the model was evaluated by MSE and R between the computed output value and objective output value of test set. RESULTS: When the value of SP was 3.0, the MSE and R values of the training set, validation set and test set were 1.35×10-5 and 0.999 85, 0.002 833 and 0.971 86, 0.005 439 and 0.936 76, respectively. RBF neural network model showed sound predictive performance and generalization ability. CONCLUSION: It is practical and valid for RBF neural network model to be applied to study steady-state plasma concentration prediction of clozapine.
展开▼
