Bos J Martijn, Liu Kan, Attia Zachi I, Noseworthy Peter A, Friedman Paul A, Ackerman Michael J
Division of Pediatric Cardiology, Department of Pediatric and Adolescent Medicine, Mayo Clinic, Rochester, MN; Department of Molecular Pharmacology and Experimental Therapeutics, Windland Smith Rice Sudden Death Genomics Laboratory, Mayo Clinic, Rochester, MN.
Division of Heart Rhythm Services, Department of Cardiovascular Medicine, Windland Smith Rice Genetic Heart Rhythm Clinic, Mayo Clinic, Rochester, MN.
Mayo Clin Proc. 2025 Feb;100(2):276-289. doi: 10.1016/j.mayocp.2024.07.016. Epub 2025 Jan 11.
To test whether an artificial intelligence (AI) deep neural network (DNN)-derived analysis of the 12-lead electrocardiogram (ECG) can distinguish patients with long QT syndrome (LQTS) from those with acquired QT prolongation.
The study cohort included all patients with genetically confirmed LQTS evaluated in the Windland Smith Rice Genetic Heart Rhythm Clinic and controls from Mayo Clinic's ECG data vault comprising more than 2.5 million patients. For the AI-DNN model, every patient and control with 1 or more ECGs above age- and sex-specific 99th percentile values for QTc (>460 ms for all patients [male/female] <13 years of age or >470 ms for men and >480 ms for women above this age) were included. LQTS patients were age and sex matched to controls at a 1:5 ratio. An AI-DNN involving a multilayer convolutional neural network was developed to classify patients.
Of the 1,599 patients with genetically confirmed LQTS, 808 had 1 or more ECGs with QTc above the defined thresholds (2987 ECGs) compared with 361,069 of 2.5 million controls (14% of Mayo Clinic patients having an ECG, "presumed negative"; 989,313 ECGs). Following age and sex matching and splitting, 3,309 (training), 411 (validation), and 887 (testing) ECGs were used. This model distinguished patients with LQTS from those with acquired QT prolongation with an area under the curve of 0.896 (accuracy 85%, sensitivity 77%, specificity 87%). The model remained robust with areas under the curve close to or above 0.9, independent of matching ratio (range, 1:5 to 1:2000) or type of ECG data used (rhythm strip of median beat) and after excluding patients with wide QRS or ventricular pacemaker.
For patients with a QTc exceeding its 99th percentile values, this novel AI-DNN functions as an LQTS mutation detector, being able to identify patients with abnormal QT prolongation secondary to an LQTS-causative mutation rather than with acquired QT prolongation. This algorithm may facilitate screening for this potentially lethal yet highly treatable genetic heart disease.
测试基于人工智能（AI）深度神经网络（DNN）的12导联心电图（ECG）分析能否区分长QT综合征（LQTS）患者和获得性QT延长患者。
研究队列包括在温德兰·史密斯·赖斯遗传性心律诊所评估的所有基因确诊的LQTS患者，以及来自梅奥诊所ECG数据库的对照，该数据库包含超过250万患者。对于AI-DNN模型，纳入每例QTc高于年龄和性别特异性第99百分位数（13岁及以下所有患者[男性/女性] >460 ms，此年龄以上男性>470 ms，女性>480 ms）且有1份或更多份ECG的患者和对照。LQTS患者与对照按1:5的比例进行年龄和性别匹配。开发了一个包含多层卷积神经网络的AI-DNN来对患者进行分类。
在1599例基因确诊的LQTS患者中，808例有1份或更多份QTc高于定义阈值的ECG（2987份ECG），而250万对照中有361,069例（梅奥诊所患者中有14%有ECG，“推测为阴性”；989,313份ECG）。经过年龄和性别匹配及分组后，使用了3309份（训练）、411份（验证）和887份（测试）ECG。该模型区分LQTS患者和获得性QT延长患者的曲线下面积为0.896（准确率85%，敏感性77%，特异性87%）。该模型在曲线下面积接近或高于0.9时仍保持稳健，与匹配比例（范围为1:5至1:2000）或使用的ECG数据类型（中位心搏节律条）无关，并且在排除宽QRS或心室起搏器患者后也是如此。
对于QTc超过其第99百分位数的患者，这种新型AI-DNN可作为LQTS突变检测器，能够识别由LQTS致病突变继发的异常QT延长患者，而非获得性QT延长患者。该算法可能有助于筛查这种潜在致命但高度可治疗的遗传性心脏病。
