基于EMD和GEP的急性低血压预测方法研究

Based on EMD and GEP acute hypotension episodes forecast methodology research

急性低血压(Acute Hypotensive Episodes,AHE)是ICU重症监护室中患者常见且危害严重的术后发症状之一。AHE的有效诊断与预测,给予医生足够时间实现干预措施,具有十分重要的临床意义。但由于血压时间序列数据高度非线性和复杂性,使得AHE的诊断与预测尤为困难。为此,面向复杂非线性时间序列的建模,本文提出一种基于经验模态分解(Empirical Mode Decomposition,EMD)和基因表达式程序设计(Gene Expression Programming,GEP)的综合方法,并构建相似性匹配模版方法来提高建模的稳定性。应用PhysioNet?中MIMIC-II的数据进行实验分析,发现本方法是有效、可行的。为复杂非线性时间序列数据的建模预测提供了一条可参考的路径。

Acute Hypotensive Episodes （AHE） is one of the recurrent postoperative symptoms occurring in intensive care units （ICU）, which always result in serious hazard of patients. Early detection and diagnosis of AHE, gives professionals enough time to select a more effective treatment, has important clinical significance. Because that the blood pressure time series data is highly nonlinear and complexity, it makes the diagnosis and detection of AHE particularly difficult. For this purpose, this paper proposes an integrated method based on Empirical Mode Decomposition （EMD） and Gene Expression Programming （GEP） for complicated nonlinear time series modeling, and build a similarity matching models method to improve the stability of the model. The methodology is applied in the context of PhysioNet MIMIC-Ⅱ（Multi-parameter Intelligent Monitoring for Intensive Care II） Database, experiments show that the methodology is effective and feasible. It is also expected that this study may offer a reference to model and forecast of the complex and nonlinear time series data.