基于BP神经网络的阿尔茨海默病预测模型研究

Self-report analysis of patients with Alzheimer's disease based on BP neural network

目的:基于阿尔茨海默病患者的日常认知自我报告清单,利用BP神经网络模型构建阿尔茨海默病不同病程的预测分类器,并评估预测分类器的性能。方法:纳入ADNI-GO、ADNI-2、ADNI-3等3个计划阶段的参与者日常认知自我报告清单数据。以7:3的比例划分训练集和测试集,以网格搜索方法设置十折交叉验证确定最佳BP神经网络参数,测试集用于评估模型的泛化能力。结果:ADNI-GO和ADNI-2阶段的模型分类准确率达到90%,而ADNI-3阶段的准确率最低,不到80%,且ADNI-3模型的泛化能力低,存在过拟合问题。结论:利用患者当前日常认知自我报告清单可以准确预测分类患者的病程阶段,有利于患者尽快进一步检查或治疗,具有一定的临床价值。

Objective The daily cognitive self-report inventory of patients with Alzheimer’s disease(AD)was used to construct a predictive classifier for different course of AD based on BP neural network model,evaluating the performance of the predictive classifier.Methods Daily cognitive self-report inventory data of participants in ADNI-GO,ADNI-2 and ADNI-3 phases were included.The data was split into training and test sets in a 7:3 ratio and the grid search method was applied to set 10-fold cross verification to determine the optimal BP neural network parameters with the test set being used to evaluate the generalization capability of the model.Results The model classification accuracy of ADNI-GO and ADNI-2 in the two planning stages reached 90%,ADNI-3,however,had the lowest accuracy,which was less than 80%.With low generalization ability of the ADNI-3 model,there was the problem of overfitting.Conclusion The disease course of the classified patient can be accurately predicted with the patient’s current daily cognitive self-report inventory,contributing to the patient’s further examination or treatment,which is of clinical significance.