BPNN在重症手足口病相关因素分析及重症化预测的应用

发布时间：2018-02-25 06:17

  本文关键词： BPNN 手足口病 重症化进程 MIV　出处：《郑州大学》2014年硕士论文　论文类型：学位论文

【摘要】：目的 该研究应用BP神经网络（BPNN）原理建立重症手足口病相关因素和重症化进程预测模型，用以探讨BPNN模型在重症手足口病临床诊断和重症化进程预测中的应用价值，为手足口病的临床诊断和流行病学研究奠定基础。 方法 以手足口病流行病学现况调查资料为基础，整群抽取河南省郑州市某医院2013年4－6月收治的344例手足口病患儿作为调查对象进行问卷调查。采用MATLAB7.0软件中的神经网络工具箱构建BPNN模型，分析得出影响重症手足口病临床诊断相关因素的平均影响值（Mean Impact Value，MIV），按MIV值的绝对值大小排出因子顺位，并与多因素logistic回归模型分析结果进行比较。对影响力较大的MIV值结果归一化得出综合因素水平计算公式，并根据收集的自发病到重症过程中有完整资料的病例，进一步分析此水平与重症化进程之间的关系。 结果 1.单因素logistic回归结果显示，精神差、血糖升高、颈强直、易惊、嗜睡、手足抖动、呕吐、肢体无力、热峰≥39℃、白细胞≥15×109/L等10个因素有意义；多因素logistic回归结果显示，易惊、手足抖动、嗜睡、呕吐、精神差、白细胞≥15×109/L、颈强直是重症手足口病临床诊断的相关因素。 2.本次训练好的网络训练样本对训练数据的分类正确率为100%，测试样本对测试数据的分类正确率＞90%，BPNN模型拟合较好。 3. BPNN模型最终网络结构设定为27→8→1，影响重症手足口病前10位相关因素（MIV值绝对值）依次为：易惊（0.4614）、精神差（0.3050）、手足抖动（0.1019）、呕吐（0.0912）、热程≥3d（0.0711）、颈强直（0.0461）、白细胞≥15×109/L（0.046）、嗜睡（0.028）、血糖升高（0.015）、呼吸节律改变（0.012）。 4.通过比较BPNN模型和多因素logistic回归结果，发现两者主要临床诊断相关因素排序顺序基本一致，热程≥3d与精神差和白细胞≥15×109/L均有交互作用（P＜0.05），热程≥3d是一个重要的协变量。 5.在重症手足口病重症化进程中，，综合因素水平在重症前一天之前的几天上升趋势明显，在重症前一天和重症当天之间略微上升，并在重症当天达到峰值，随后下降。 结论 BPNN模型可用于建立重症手足口病相关因素模型，并可对手足口病新发病例作出重症化进程预测，可用于手足口病的临床诊断和重症化进程预测。
[Abstract]:Purpose. In this study, BP neural network (BP neural network) principle was used to establish the prediction model of severe hand, foot and mouth disease (HFMD) related factors and severity process, and to explore the application value of BPNN model in the clinical diagnosis and prognosis of severe HFMD. To lay a foundation for the clinical diagnosis and epidemiological study of hand, foot and mouth disease. Method. Based on the epidemiological data of hand, foot and mouth disease, A cluster survey was conducted on 344 children with hand, foot and mouth disease in a hospital in Zhengzhou, Henan province from April to June in 2013. The BPNN model was constructed by using the neural network toolbox of MATLAB7.0 software. The results showed that the mean value of influencing factors related to the clinical diagnosis of severe hand, foot and mouth disease was the mean value of mean Impact value, and the rank of factors excreted according to the absolute value of MIV value. The results were compared with the results of multivariate logistic regression model. The formula of comprehensive factor level was obtained by normalizing the results of the influential MIV values, and according to the collected cases with complete data from the onset to the critical stage, Further analysis of the relationship between this level and the severity process. Results. 1. The results of univariate logistic regression showed that 10 factors were significant, such as mental retardation, increased blood glucose, neck rigidity, agitation, drowsiness, wobble of hands and feet, vomiting, limb weakness, heat peak 鈮

本文编号：1533320