Application Of Multi-parameter Fusion In Spontaneous Circulation Recovery Identification And Stability Assessment During Cardiopulmonary Resuscitatio

Objective:It is difficult to accurately identify restoration of spontaneous circulation(ROSC)in patients with cardiac arrest(CA)by single physiological parameter of electrocardiograph(ECG),pulse oximetry photoplethysmogram(POP)or end-tidal carbon dioxide(ETCO2).Some patients with ROSC cannot maintain sustained ROSC,and non-sustainable ROSC is an important obstacle to the survival of CA patients.This study explored the system of multi-parameter fusion for early and accurate identification of ROSC in experimental animals through a machine learning approach.An attempt was made to evaluate whether ETCO2 and POP parameters and their combination could be used to identify unsustained ROSC.Methods:Pigs were used as models to simulate the CA patients.The ECG,ETCO2,and POP were recorded.The characteristics of ECG,POP and ETCO2 were extracted.The Random Forest model is applied for feature selection,and then trained using models such as K-Nearest Neighbor,Support Vector Machine,Naive Bayes,Decision Tree,Logistic Regression,Neural Network,Random Forest and Adaboost.The identification efficacy of each model on ROSC and the time before the clinician identified ROSC were obtained.We carried out a multi-central observation cohort study for patients with CA from 2013 to 2014.The general information of the patient was collected and the parameters of ECG,POP,and ETCO2 were analyzed.Results:1.This study suggests that Support Vector Machine(SVM),Neural Network(NN),Random Forest(RF)and AdaBoost model can reach 100%positive prediction value of ROSC.2.This study suggested that the area under curve of ROC of Support Vector Machine(SVM),Neural Network(NN),Random Forest(RF),Logistic Regression(LR)and AdaBoost model for the identification of ROSC during cardiopulmonary resuscitation were respectively 0.77,0.80,0.84,0.79 and 0.81 based on the multiparameter fusion of three parameters(POP+ECG+ETCO2).3.RF model(three-parameter fusion)has significantly higher area under ROC than other machine learning models in predicting ROSC.4.The included 105 ROSC episodes(from 80 cardiac arrest patients)comprised 51 sustained ROSC episodes and 54 unsustained ROSC episodes.Patients with unsustained ROSC had lower 24-hour survival rate than sustained ROSC patients(11.5%vs.25.0%,respectively;P<0.05).5.The logistic regression analysis showed that,the difference between after and before ROSC in ETCO2(△ETCO2)and the difference between after and before ROCS in area under the curve of pulse oximetry photoplethysmogram(△AUCp)were independently associated with sustained ROSC(OR=0.931,P=0.011 and OR=0.998,P<0.001).6.The area under the receiver operating characteristic curve of △ETCO2,△AUCp,and the combination of both to predict unsustained ROSC were 0.752(95%CI 0.6600.844),0.883(95%CI 0.818-0.948),and 0.902(95%CI 0.842-0.962).Conclusion:1.The model of SVM,NN,RF and AdaBoost adopted the multi-parameter fusion of three physiological parameters(ECG,POP,ETCO2)had high efficacy for the identification of ROSC during cardiopulmonary resuscitation;The effectiveness of RF model is significantly better than other machine learning models.2.Patients with unsustained ROSC have a poor prognosis.3.Combined △ETCO2 and △AUCp showed significant predictive value for unsustained ROSC.