一种新的特征提取方法在无创动脉血压信号异常识别中的应用

A Novel Feature Extraction Method for Signal Anomaly Identification of Non-invasive Arterial Blood Pressure for Monitoring Cerebral Autoregulation

目的:脑血流自动调节机能是指动脉血压在一定范围内变动时,脑血流量维持恒定的生理调节机能。目前缺乏对其输入信号(无创连续动脉血压信号)的异常检测算法,本文提出一种新的特征提取方法并应用于该信号的异常检测。方法:本文针对有创血压信号异常检测算法(SAI算法)的缺陷提出两个简单而有效的特征,舒张末期斜率和、慢速射血期斜率和,并采用累积概率分布函数(CDF)和接收机工作特性分析(ROC),挑选最佳特征并得到其最佳参数。将SAI算法和新特征结合,用于无创连续动脉血压信号的异常识别。最后应用临床实际采集到的数据进行了效果验证。结果:实验表明,该算法的敏感度可达93.95%,特异度可达84.87%。结论:本文方法具有结构简单、易实现、运算量小等优点,可有效筛选出异常信号,为稳定地监测动态血流调节机能提供高质量的血压输入信号。

Objective：Cerebral autoregulation is arterial blood pressure changes in a certain range,the blood flow to maintain constant physiological control mechanism.The current lack of its input signal（non-invasive continuous arterial blood pressure signal） of the anomaly detection algorithm.This paper presents a new method of feature extraction and anomaly detection ap-plied to the signal.Methods：According to the defect of the invasive blood pressure signal anomaly detection algorithm（SAI algorithm）,we proposed two feature both simple and effective,end-diastolic slope sum,and slow ejection of the slope sum.And using cumulative probability distribution function（CDF） and receiver operating characteristic analysis（ROC）,select the best features and get the best parameters.SAI algorithms and new features will be combined,for noninvasive arterial blood pressure signal of anomaly identification.Finally the actual clinical data were collected to verify the performance.Results：The results show that the algorithm can reach 93.95% sensitivity and 84.87% specificity.Conclusions：This method has a simple structure,easy implementation,computational complexity,etc,and can effectively screen out the abnormal signal to provide high-quality the blood pressure input signal in order to monitor the dynamic flow control mechanism stabilized.