关键词: 闭塞性心肌梗死, 心电图, 人工智能, 风险分层, 重庆市

Abstract: Background　The occurrence rate of occlusive myocardial infarction (OMI) continues to rise. However，there is currently a lack of reliable and accurate diagnostic tools during the initial clinical diagnosis stage. There is an urgent need for accurate and timely diagnostic tools to enable early diagnosis and treatment，thereby preventing disease progression and improving prognosis. Objective　This study aims to develop and internally validate a model based on 12-lead ECG to predict the probability of angiography-confirmed OMI, and to derive risk stratification for emergency triage. Methods　We utilized 18 259 matched ECG and digital subtraction angiography datasets from 17 071 patients. A total of 51 ECG features were extracted from each dataset. Thirteen different models were compared in terms of classification performance. Feature selection was performed using t-tests and Pearson correlation analysis to compare classification outcomes across different feature subsets. Based on the optimal model，an OMI risk assessment model was constructed according to the probability density distribution of the disease，enabling risk stratification for OMI. Results　The Gradient Boosting Decision Tree (GBDT) model achieved the best performance on the test set when using all features，with an area under the receiver operating characteristic curve (AUROC) of 0.899，outperforming human ECG experts. Furthermore，the OMI risk score derived from this model improved both the inclusion and exclusion accuracy，achieving a specificity of 0.952 for high-risk OMI. The sensitivity for the low-risk category is 0.849，with a negative predictive value (NPV) of 0.987 and a miss rate of 1.33%. Conclusion　We developed an AI-ECG diagnostic algorithm for OMI detection and constructed a risk assessment model based on this algorithm. This model holds promise for application in emergency triage，assisting clinicians in diagnosing and risk-stratifying OMI patients more rapidly and accurately.

Key words: Occlusive myocardial infarction, Electrocardiogram, Artificial intelligence, Risk stratification, Chongqing