Evaluation Of Ecg Changes In Patients With Cerebral Infarction Based On Artificial Intelligence

Research background:Cerebral infarction is the most common type of stroke,accounting for about70% ~ 80% of all strokes.Studies have shown that the second leading cause of death after stroke is cardiovascular complications.67% of patients with acute cerebral infarction have abnormal ECG of ischemia and / or arrhythmia within 24 hours of onset.Within 3 months after the onset of cerebral infarction,about 4% of patients died of cardiac causes,and about 19% of patients had fatal or serious non fatal cardiac events.ECG monitoring is effective in improving the detection and treatment of heart damage after stroke.Some current guidelines even recommend ECG monitoring for at least 24 hours after stroke.The American College of Cardiology(ACC)and the American Heart Association(AHA)reviewed and analyzed ECG articles and found that 33% of ECG had differences in interpretation.Even experienced cardiologists or experts may have different diagnoses.Patients with undetected "pseudo normal" ECG will be delayed and even life-threatening.Objective:Adopting deep learning method,by constructing machine learning model,which can quickly and automatically identify the ECG changes of patients with cerebral infarction.The model’s results will evaluate patient’s conditions,reducing the mortality of patients with cerebral infarction caused by cardiovascular complications and improving the prognosis.Methods:In this study,the standard 12 lead ECG results of patients with first-episode cerebral infarction from September 2017 to May 2020 were obtained from a hospital’s big data center in Nanchang.They were randomly divided into training set,verification set and test set in the ratio of 6:2:2.Models including logistic regression,random forest(RF),support vector machine(SVM),BP neural network(BP-ANN)and convolutional neural network(CNN)were constructed form the training set,and the training model was cross verified with the verification set.The test set was used to evaluate the performance of the models in identifying the changes of ECG signals in patients with cerebral infarction.Results:(1)A total of 6732 electrocardiograms were introduced in this study,including3367 electrocardiograms of patients with cerebral infarction,of which 771 cases(22.89%)were diagnosed as normal electrocardiograms by electrocardiographists,and 2596 were diagnosed as abnormal electrocardiograms and 3365 cases of normal sinus rhythm electrocardiograms form patients in other departments;(2)According to the AUC value from high to low,the evaluation effects of the four machine models according to the traditional ECG indexes of ECG recognition are as follows: BP neural network(0.775,95% CI: 0.752 ~ 0.796),logistic regression(0.705,95% CI: 0.681 ~ 0.729),random forest(0.686,95% CI: 0.685 ~ 0.688),SVM(0.647,95% CI: 0.622 ~ 0.673);(3)The performance of CNN model in identifying 12 lead ECG changes in patients with cerebral infarction was higher than the other machine models described above.Its AUC value,the sensitivity,the specificity and the accuracy was 0.885(95% CI,0.865-0.903),78.6%,84.2%,and 81.9% respectively;(4)The prognosis of 331 patients with cerebral infarction was evaluated by CNN.The modified Rankin Scale score showed 167 cases in the poor prognosis group and164 cases in the good prognosis group.The results of CNN model showed that there was significant difference in ECG scores between the two groups(P < 0.05)。Conclusions:(1)Convolution neural network has more advantages in identifying the changes of ECG in patients with cerebral infarction than the traditional ECG indexes recognized by ECG machine;(2)The ECG at admission is in good agreement with the prognosis after discharge,suggesting that ECG may predict the prognosis of patients with cerebral infarction.