探索应用人工智能辅助心电图识别拟诊冠心病患者中的无冠心病人群（冠状动脉狭窄程度<50%）。

本研究是一项基于回顾性心电图数据建立预测无冠心病人群人工智能模型并加以验证的临床研究。入选天津医科大学第二医院、天津市第三中心医院和厦门长庚医院门急诊2020年1月至2022年12月期间的拟诊冠心病并于住院期间行冠状动脉造影检查的患者，基于标准12导联心电图建立心电图数据集。按照主要冠状动脉或其主要分支的狭窄程度是否<50%，将心电图标记为无冠心病组和阳性对照组。通过训练心电图建立深度神经网络模型，识别无冠心病人群。

共纳入4 489例病历，其中4 187例用于模型构建，302例厦门长庚医院数据用于模型的外部验证。模型内部验证的接收者操作特征曲线下面积（AUC）为0.70，敏感性为0.701，特异性为0.630，F1评分为0.469。外部验证的AUC为0.55，敏感性为0.359，特异性为0.784，F1评分为0.373。

基于心电图的人工智能模型能够识别拟诊冠心病患者中的无冠心病人群，具有一定的临床应用价值。

To explore the application of artificial intelligence-assisted electrocardiograms (ECG) to identify patients without coronary heart disease (coronary artery stenosis < 50%).

Patients with suspected coronary heart disease and who underwent coronary angiography during hospitalization were enrolled. The ECG data set was established based on standard 12-lead ECG. The ECG was labeled as a group without coronary disease and a control group according to whether the main coronary artery or its main branches were narrowed by less than 50% in diameter. A deep neural network model was established by training ECG to identify patients without coronary heart disease.

A total of 4 489 ECG medical records were included, of which 4 187 were used for model construction and 302 were used for external verification of the model. The area under curve (AUC) value, sensitivity, and specificity of the model were 0.70, 0.701, 0.630, and 0.469 respectively. The AUC value of external validation was 0.55, sensitivity 0.359, specificity 0.784, and F1 score 0.373.

The artificial intelligence model based on ECG can identify patients without coronary heart disease in patients with suspected coronary heart disease, which has certain clinical application value.