目的应用深度学习算法建立基于纸质版12导联体表心电图的人工智能模型,评估并比较其与心律失常专科医生对流出道室性心律失常起源部位定位的效能。
方法收集2011年5月1日至2020年12月31日在首都医科大学附属北京朝阳医院、温州医科大学附属第二医院、浙江大学医学院附属邵逸夫医院、北京大学第三医院经射频消融术成功治疗的527例流出道室性心律失常患者的861份纸质版室性心律失常心电图。以术中成功消融靶点作为金标准,对以深度学习为基础的人工智能模型进行训练、验证及测试。两位心律失常专科医生根据经典定位流程图及临床经验对测试集心电图进行靶点判断。对比深度学习算法与人工判读的敏感性、特异性和准确性等。
结果基于纸质版常规12导联体表心电图的扫描图像建立的人工智能模型的灵敏度、特异度、准确率、受试者工作特征曲线下面积、F1指数分别为94.0%、83.3%、93.0%、0.93和0.96。针对测试集,专科医生判读的平均灵敏度、特异度和准确率分别为95.7%、83.3%和94.6%。
结论在流出道室性心律失常起源定位方面,基于纸质版心电图的深度学习算法表现出较高性能,与心律失常专科医生的判断水平相当。
ObjectiveTo develop establish an artificial intelligence model based on paper 12 lead body surface electrocardiogram(ECG) by using deep learning (DL) algorithm, and to evaluate and compare its effectiveness with arrhythmia specialists in locating the origin of outflow tract ventricular arrhythmia.
MethodsA total of 861 paper ECG were collected from 527 patients who were successfully treated with radiofrequency ablation for OT-VA in Beijing Chao-Yang Hospital, The 2nd Affiliated Hospital of Wenzhou Medical University, Sir Run Run Shaw Hospital of Zhejiang University College of Medicine and Peking University Third Hospital from May 1, 2011 to December 31, 2020.Taking the successful ablation target as the gold standard, the artificial intelligence model based on DL was trained, verified and tested.Two electrophysiologists judged the target according to the classic localization flow chart and clinical experience and compared its sensitivity, specificity, accuracy with the DL algorithm.
ResultsThe DL algorithm achieved an area under the receiver operating characteristic curve ( AUC) of 0.93 with sensitivity, specificity, accuracy and F1 scores of 94.0%, 83.3%, 93.0% and 0.96 respectively, while the electrophysiologists identified the ventricular arrhythmia chamber of origin with a sensitivity of 95.7%, specificity of 83.3% and accuracy of 94.6%.
ConclusionIn terms of the origin location of outflow tract ventricular arrhythmia, the DL algorithm shows high performance, which was comparable to the judgment level of arrhythmia specialists.
周杨,章德云,魏国栋,等. 基于纸质版心电图应用深度学习算法定位流出道室性心律失常起源部位的研究[J]. 中华心律失常学杂志,2022,26(02):127-131.
DOI:10.3760/cma.j.cn113859-20220214-00026版权归中华医学会所有。
未经授权,不得转载、摘编本刊文章,不得使用本刊的版式设计。
除非特别声明,本刊刊出的所有文章不代表中华医学会和本刊编委会的观点。
组别 | 例数 | 年龄(岁,
|
男[例(%)] | BMI(kg/m
2,
|
LVEDD(mm,
|
LVEF(%,
|
局部电位较体表QRS波提前时间(s,
|
|
---|---|---|---|---|---|---|---|---|
训练集 | RVOT | 300 | 48.79±15.94 | 107(35.7) | 25.18±4.62 | 47.73±4.01 | 65.94±7.12 | 35.03±7.13 |
LVOT | 48 | 51.86±13.50 | 15(31.3) | 27.00±4.30 | 47.06±4.73 | 66.82±10.55 | 32.51±7.27 | |
P值 | 0.329 | 0.552 | 0.073 | 0.476 | 0.707 | 0.126 | ||
验证集 | RVOT | 41 | 42.43±17.16 | 11(26.8) | 27.36±3.89 | 49.46±6.91 | 66.79±10.05 | 30.89±7.32 |
LVOT | 9 | 39.76±14.96 | 4(44.4) | 25.13±4.28 | 35.67±5.47 | 64.67±5.35 | 31.50±8.83 | |
P值 | 0.200 | 0.296 | 0.249 | <0.001 | 0.622 | 0.860 | ||
测试集 | RVOT | 117 | 45.02±13.15 | 42(35.9) | 34.07±4.21 | 47.08±4.84 | 68.14±5.93 | 34.70±6.64 |
LVOT | 12 | 58.13±10.54 | 7(58.3) | 27.39±5.37 | 51.52±5.40 | 62.23±14.47 | 23.85±7.89 | |
P值 | 0.023 | 0.127 | 0.142 | 0.041 | 0.033 | 0.006 |
注:RVOT为右心室流出道,LVOT为左心室流出道,BMI为体重指数,LVEDD为左心室舒张末期内径,LVEF为左心室射血分数

你好,我可以帮助您更好的了解本文,请向我提问您关注的问题。