误差逆向传播神经网络模型对脑出血患者血肿体积扩大的预测作用

Predictive effect of back propagation neural network model on hematoma enlargement in patients with cerebral hemorrhage

目的研究误差逆向传播(BP)神经网络模型对脑出血患者血肿扩大(HE)的预测结果。方法回顾性分析解放军第三○九医院于2011年1月至2014年12月连续入院并诊断为脑出血的128例患者的临床资料。应用Matlab 7.14软件实现BP神经网络模型对脑出血患者24 h内HE(HE≥6.0 ml和HE≥12.5 ml)的预测。计算模型的均方差(MSE)以及整体预测的准确度;绘制预测HE的受试者工作特征(ROC)曲线。结果 BP神经网络预测HE≥6.0 ml和HE≥12.5 ml时,训练集、校验集、测试集的MSE分别为0.061、0.143、0.052和0.023、0.057、0.065。对HE拟合最佳验证性能分别为:HE≥6.0 ml网络训练11次,误差值为0.224;HE≥12.5 ml网络训练20次,误差值为0.057。预测HE≥6.0 ml和HE≥12.5 ml的整体准确度分别为92.2%(118/128)、96.9%(124/128)。结论 BP神经网络模型对数据要求无特殊限制,能够较为准确地拟合脑出血HE模型。

Objective To study predicting results of the back propagation（ BP） neural network model for hematoma enlargement（ HE） in patients with intracerebral hemorrhage. Methods The clinical data of 128 patients with cerebral hemorrhage admitted to the 309 th hospital of People＇ s Liberation Army from January 2011 to December 2014 were analyzed retrospectively. The Matlab 7. 14 software was used to achieve BP neural network model for predicting hematoma enlargement within 24 hours in patients with intracerebral hemorrhage（ HE ≥6. 0 ml and HE ≥12. 5 ml）. The mean square error（ MSE） of the model and the accuracy of the overall prediction were calculated. The receiver operation characteristic（ ROC）curve was drawn for predicting HE. Results When the BP neural network predicted HE ≥6. 0 ml and HE ≥12. 5 ml,the mean square deviations of the training set,validation set,and test set were 0. 061,0. 143,0. 052 and 0. 023,0. 057,and 0. 065,respectively. The best fitting performance verification of hematoma enlargement was as follows： ≥ 6. 0 ml for network training 11 times and the error value0. 224; ≥12. 5 ml for network training 20 times,and the error value 0. 057. The overall accuracies of predicting HE ≥6. 0 ml and HE ≥12. 5 ml were 92. 2%（ 118 /128） and 96. 9%（ 124 /128） respectively.Conclusion The BP neural network model have no special limitation for data. It can accurately fit the hematoma expansion model of cerebral hemorrhage.