Application Of Order Statistics Information In Defibrillation Rhythm Recognition

According to the "China Cardiovascular Disease Report 2018" issued by the National Cardiovascular Center,the number of sudden cardiac deaths in China exceeds 540,000 per year,which is equivalent to about 1,500 deaths per day due to sudden cardiac death.Cardiac sudden death is mainly caused by Ventricular Fibrillation(VF)and Ventricular Tachycardia(VT).There is often no warning of VF/VT episodes.The electrical activity of the ventricle is not synchronized during the attack,and the heart pumping function is lost.If the measures are not taken in time to transfer the heart rhythm,the patient will die within a few minutes.The deployment of Automatic External Defibrillator(AED)in public places with high traffic is an effective means of timely rescue of VF/VT patients outsid e the hospital.The key technologies of AED are Shockable Rhythms(Sh R)(including VT and VF)and Non-shockable Rhythms(Nsh R)(including sinus arrhythmia,atrial flutter,atrial fibrillation,ventricular escape).Accurate identification of beats,etc.).Rapid and accurate diagnosis of Sh R and Nsh R is important for the rescue of patients with cardiac arrest.However,existing methods require longer ECG signals to accurately identify Sh R/Nsh R.The reasons are as follows: 1)The extracted features fail to met iculate the statistical distribution of the ECG waveform;2)The classifier does not utilize the timing information of the ECG features.3)The feature extraction function is complex and has too many parameters,which may not be optimal.In response to these problems,we propose a sequential neuralization of the Convolutional Neural Network(CNN)to filter the ECG signals and extract the sequential statistics as the distribution characteristics of the ECG waveforms;then CNN and the length of time The memory neural network cascades to capture dynamic information of the distribution features to identify Sh R/Nsh R.Five-fold cross-validation evaluation was performed on the open ECG datasets CUDB,VFDB and AHADB.Our method uses a 2-second ECG signal classification and the equalization error rate is 1.87%,while the existing three schemes use the 8-second ECG signal to obtain 2.83% respectively.,2.98%,3.05% balanced error rate.The results show that the 2 second signal recognition error rate of this scheme is lower than the traditional method for 8 seconds,and the detection of ECG signal duration is shortened by 4 times.The lower recognition error will reduce the occurrence of misdiagnosis and missed diagnosis,and win valuable for the rescue of out-of-hospital cardiac arrest.Time and vitality.