【摘要】 目的 基于阳性反应心电图，应用人工智能技术创建血管迷走性晕厥（VVS）识别模型。方法 回顾性收集2023年1月1日至2024年6月30日在郑州大学第一附属医院进行直立倾斜试验检查的病例资料共2278例。根据直立倾斜试验检查结果及临床病史，分为VVS组（1048例）和对照组（1230例），收集患者年龄、性别、基础血压、基础心率、心电图原始数据等资料，心电图原始数据进一步分为平卧位心电图、体位变化心电图、倾斜位心电图及VVS阳性反应时心电图，两组患者分别通过简单随机法按3∶1∶1分为训练集、验证集和测试集，并进行VVS识别模型训练。结果 VVS组和对照组年龄分别为（43.55±19.33）岁和（41.78±19.84）岁，男性比例分别为38.84%和53.52%，收缩压分别为（125.87±18.06）mmHg和（129.14±18.28）mmHg，舒张压分别为（78.29±11.00）mmHg和（79.74±11.48）mmHg，心率分别为（73.59±12.09）次/min和（74.75±12.88）次/min，均存在统计学差异（P均<0.05）。经过训练、验证和测试，基于平卧位心电图、体位变化心电图、倾斜位心电图所建立的VVS识别模型F1分数均为0.53,AUC均为0.59，基于VVS阳性反应时心电图所建立的VVS识别模型F1分数为0.78,AUC均为0.89。结论 本研究基于VVS阳性反应时心电图，应用人工智能技术成功构建了VVS识别模型，为VVS的诊断和临床干预提供了新的方法和工具，有望改善患者的诊疗效果。更多还原

【Abstract】 Objective To develop a vascular vagal syncope(VVS) recognition model by artificial intelligence technology based on the VVS positive reaction of electrocardiogram. Methods A retrospective collection of case data was performed for 2278 patients who underwent head-up tilt tests at the First Affiliated Hospital of Zhengzhou University from January 1,2023, to June 30,2024. According to the results of head-up tilt tests and clinical history, the cases were divided into the VVS group(1,048 cases) and the control group(1,230cases). Data such as the age, gender, baseline blood pressure, baseline heart rate and raw electrocardiogram data were collected. The raw electrocardiogram data were divided into the supine position, position changing, oblique position and VVS positive reaction. Two groups of patients were respectively divided into training set, validation set and test set according to the ratio of 3∶1∶1 by simple random method, and the VVS recognition model was trained. Results The mean age of the VVS group and the control group were(43.55±19.33) years and(41.78±19.84) years, with the male proportion 38.84% and 53.52%, the systolic blood pressure(125.87±18.06) mmHg and(129.14±18.28) mmHg, the diastolic blood pressure(78.29±11.00) mmHg and(79.74±11.48) mmHg, the heart rate(73.59±12.09) beats/min and(74.75±12.88) beats/min, respectively; and all the statistically significant differences(all P<0.05). After training, validation and testing, the F1 score and AUC of VVS recognition models based on the electrocardiograms of supine position, position changing and oblique position were all 0.53 and 0.59, respectively.The F1 score and AUC of VVS recognition models based on electrocardiogram of VVS positive reaction were 0.78and 0.89, respectively. Conclusions Based on the VVS positive reaction of the electrocardiogram, the VVS recognition model is successfully constructed using artificial intelligence technology, providing new methods and tools for the diagnosis and clinical intervention of VVS, which is expected to improve the diagnosis and treatment effectiveness of VVS patients.更多还原