The Prediction Of Acute Hypotension Episode In The Intensive Care Unit

Cardiovascular disease is one of the common diseases. Specifically acute hypotension especially in Intensive Care Unit is a very dangerous sudden emergency, and serious threaten human’s safety of life. Patients in ICU usually have a very unstable physical condition; Health care needs often appear intervention in emergency situationsDue to limit medical resources, and acute hypotension belongs sudden emergency, it’s difficult to directly observe the occurrence sign, So AHE is a serious negative impact on patient survival. Therefore, how to detect and predict early acute hypotension has already become a clinical problem in the medical profession. Basis in the current massive clinical databases, it has significance mean to research variation of sudden emergency occurs using applications of data mining analysis and research methods combine the computer, signal processing, artificial intelligence and clinical data. Thereby find ways to predict the occurrence of sudden emergency, and then design the intelligent early warning software. Based on this, the following research carried out:1. Study processes of physiological parameters signals such as heart rate, systolic blood pressure, diastolic blood pressure, mean arterial pressure, pulse and oxygen using signal analysis, noise removal, handling missing values and outliers, extraction of statistical characteristics and characteristics of energy from multi-parameter physiological and relationships between each feature and AHE, obtained50characteristics with strong association, and select12features from50to build research eigenvectors;2. Construction two neural network models:the Levenberg-Marquardt neural network and Multiple-Output Chebshev-Polynomial Chebyshev neural network. Predict acute hypotensive episodes prediction with multiple physiological parameters as input feature vectors, Since initial weights assigns randomly, lead to network instability, in response to this phenomenon, we use ten-fold cross-validation method. comparative analysis the two models in three performance indicators of accuracy, sensitivity, specificity, Achieved good results and the best model to predict the classification;3. Proposed a method to predicting acute hypotension based on the pulse wave transit time. Studied the linear relationship between pulse wave transit time and blood pressure. Designed the pulse wave signal extraction method based on wavelet transform and threshold settings, and extract energy characteristics of the statistical characteristics of the pulse wave transit time. Studied the relationship between the onset of the pulse wave transit time of each feature and acute hypotension based on mathematical statistical methods to identify which features has strong correlation. Establish acute hypotensive episodes prediction model based on LM neural network. The results prove that pulse wave transit time can be used to predictive of acute hypotensive episodes;This research aims to study the feasibility and new ways of acute hypotensive episodes prediction mode, provide a theoretical reference and technical support for research applications in acute hypotensive episodes prediction research.