Advantages of Artificial Intelligence in Predicting Risk Factors for Stroke: Model Development and Retrospective Data Analysis (Preprint)

BACKGROUND Cerebral stroke is a common cardiovascular disease in the field of neurology. Current imaging detection methods and psychological nerve scoring methods are characterized by low sensitivity and high subjectivity. Machine learning in artificial intelligence (AI) systems has shown high accuracy in the diagnosis and treatment of diseases and has been applied in the field of neurology. At present, there are few studies on the application of machine learning to stroke diagnosis. OBJECTIVE The aim of this study was to explore the predictive value of AI systems in stroke disease and to provide a reference for the application of AI systems in the field of medical neurology. METHODS A retrospective analysis was performed on 763 patients with stroke, of whom 183 (24.0%) had recurrent stroke, as confirmed by the neurology department of First Affiliated Hospital of Zhengzhou University from January 2014 to December 2019. Basic data and medical data of all participants were collected. Univariate and multivariate Cox and logistic regression model algorithms were respectively used to predict stroke risk factors. The receiver operating characteristic (ROC) curve was used to detect the accuracy and sensitivity of the Cox and logistic models. According to the support vector machine (SVM) algorithm in machine learning, data were filled and preprocessed using mean value, median, linear regression, and normalized expectation-maximization (EM) methods. The influencing factors were selected using the conservative mean method, and the risk factors for stroke recurrence were predicted by the SVM model. The area under the curve (AUC) of the ROC curve was used to analyze and compare the prediction results of the three models. RESULTS The multivariate Cox model and logistic model analyses showed that family history of stroke, systolic blood pressure, history of heart disease, total cholesterol, disease progression, dietary habits, and history of hypertension were the main risk factors for stroke recurrence. The sensitivity and specificity of the Cox model were 0.754 and 0.805, respectively. The AUC of the logistic model was 0.889. In the SVM model data filling algorithm, the median AUC was 0.874, which was significantly higher than that of the other algorithms (P<.05). The top 10 risk factors of stroke patients predicted by the SVM model included both clinically established risk factors and potential risk factors. The prediction results of stroke risk factors were in the order of 0.873SVM>0.861logistic>0.853Cox. CONCLUSIONS AI systems have obvious advantages in the prediction of stroke disease, and this work provides a reference for the application of AI in the field of medical neurology.