基于BiLSTM-Attention模型的缺血性脑卒中的年卒中风险预测  

Risk prediction of annual stroke of ischemic strokebased on BiLSTM-Attention model

作　　者：骆轶姝[1] 邵圆圆 陈德华[1] LUO Yishu;SHAO Yuanyuan;CHEN Dehua(School of Computer Science and Technology,Donghua University,Shanghai 201620,China)

机构地区：[1]东华大学计算机科学与技术学院,上海201620

出　　处：《东华大学学报（自然科学版）》2021年第4期62-68,共7页Journal of Donghua University(Natural Science)

基　　金：上海市经信委人工智能创新发展专项资金资助项目(RX-RJJC-08-16-0483,2017-RGZN-01004)。

摘　　要：为实现缺血性脑卒中的年卒中风险的预测,提出一种基于BiLSTM(bi-directional long short-term memory)-Attention的预测模型。采用BiLSTM对患者诊断数据进行特征学习以捕获前向和后向序列数据中的信息;增加Attention机制,对隐藏层中指标信息进行权重分配来提高关键信息的有效利用率;数据集构建充分考虑缺血性脑卒中发生的影响因子,选取Logistic统计学分析方法进行确定,并定义一年为脑卒中研究时间周期,以患者当前及回归预测一年后的参数作为预测模型输入指标。试验结果表明,相比较单一模型,时序特征的提取融合Attention机制的BiLSTM模型的方法,在准确度、灵敏度、特异度等临床判断标准下的预测风险的效果较优,准确率达86%,在缺血性脑卒中疾病早筛查、早预防等领域具有良好的应用价值。A prediction model based on BiLSTM(bi-directional long short-term memory)-Attention was proposed to achieve the prediction of annual stroke risk in ischemic stroke.The BiLSTM was used to learn the features of patient diagnostic data to capture the information in the forward and backward sequential data.The Attention mechanism was added to increase the effective utilization of key information by assigning weights to the indicator information in the hidden layer.The influencing factors of the occurrence of ischemic stroke determined by the Logistic statistical analysis method were fully considered when constructing the data set,and one year was defined as stroke study time period to use the current index of patients and the parameters one year after the regression prediction as the input of the prediction model.The experimental results show that compared with the single model,the method of BiLSTM model fused with Attention mechanism in the extraction of time-series features has better results in predicting risk with 86%accuracy under the clinical judgment criteria of accuracy,sensitivity and specificity,and has good application value in the fields of early screening and prevention of ischemic stroke diseases.

关 键 词：缺血性脑卒中 年卒中风险 BiLSTM Attention机制 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] R743.3[自动化与计算机技术—控制科学与工程]