ICU急性低血压预测方法研究

发布时间：2018-07-25 20:09

【摘要】：急性低血压是重症监护室(Intensive Care Unit, ICU)病人经常出现的突发急症之一,如果不采取及时有效的干预措施会严重威胁病人的生命安全。由于目前ICU普遍面临医疗资源紧张、人手严重短缺的问题,并且急性低血压发作前缺少可以直接观察到的征兆,医护人员可能无法及时发现急性低血压发作的病人,导致病人存活率下降。因此预测急性低血压的发生或者甄选急性低血压发作高风险病人是ICU监护迫切需要解决的临床问题之一。利用ICU监护产生的海量临床数据,借助计算机自动分析、挖掘这些临床数据蕴含的急性低血压发作特征模式,实现急性低血压发作的智能预测,是解决这个问题的思路之一。基于此,本文开展了以下研究工作： 1、研究了急性低血压发作前后心率、动脉收缩压、动脉舒张压、动脉平均压、脉搏、血氧多个生理参数的变化规律,采用相关性分析方法确定了预测急性低血压发作的特征向量； 2、设计了基于LM算法的人工神经网络和多输出切比雪夫神经网络两种模型实现了急性低血压发作的预测。并将两种模型与经典BP神经网络的性能指标进行对比分析； 3、根据脉搏波传导时间与动脉血压具有相关性的特点提出了基于脉搏波传导时间的特征提取方法。分析了急性低血压发作前后脉搏波传导时间的统计特征和能量特征,采用相关性分析和主成份分析方法构建了特征向量,并采用基于LM算法的神经网络实现急性低血压发作的预测。本文研究旨在研究基于ICU临床监护数据、模式识别和人工智能技术的急性低血压发作预测方法,研究结果表明论文提出的方法取得较好的预测结果,可以为急性低血压发作预测的临床应用提供理论参考。
[Abstract]:Acute hypotension is one of the emergent emergencies frequently occurring in (Intensive Care Unit, ICU) patients in intensive care unit (ICU). If no timely and effective intervention is taken, the life safety of patients will be seriously threatened. At present, ICU is generally faced with the problems of shortage of medical resources, severe shortage of manpower, and the lack of directly observed signs before acute hypotension, so health care workers may not be able to detect patients with acute hypotension in time. This leads to a decline in patient survival. Therefore, predicting the occurrence of acute hypotension or selecting patients at high risk of acute hypotension is one of the urgent clinical problems in ICU monitoring. It is one of the ways to solve this problem to mine the characteristic pattern of acute hypotension by using the massive clinical data generated by ICU monitoring and with the help of computer automatic analysis to realize the intelligent prediction of acute hypotension. Based on this, the following research work was carried out: 1. The changes of heart rate, arterial systolic pressure, arterial diastolic pressure, mean arterial pressure, pulse and blood oxygen were studied before and after acute hypotension. The characteristic vectors for predicting acute hypotension were determined by correlation analysis. 2. Two models of artificial neural network based on LM algorithm and multiple output Chebyshev neural network are designed to predict acute hypotension. The two models are compared with the classical BP neural network. 3. According to the correlation between pulse wave conduction time and arterial blood pressure, a feature extraction method based on pulse wave conduction time is proposed. The statistical and energy characteristics of pulse wave conduction time before and after acute hypotension were analyzed. The characteristic vectors were constructed by correlation analysis and principal component analysis. A neural network based on LM algorithm is used to predict acute hypotension. The purpose of this study is to study the prediction method of acute hypotension based on ICU clinical monitoring data, pattern recognition and artificial intelligence. The results show that the proposed method has good prediction results. It can provide theoretical reference for clinical application of predicting acute hypotension.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R459.7;TP183

【参考文献】