Ning Hongxia, Yu Qi, Kong Fanxin, Li Guangyuan, Wang Yonghuai, Zhang Bo, Feng Yong, Ma Chunyan
Department of Cardiovascular Ultrasound, the First Hospital of China Medical University, Shenyang, China.
Clinical Medical Research Center of Imaging in Liaoning Province, Shenyang, China.
Quant Imaging Med Surg. 2025 Apr 1;15(4):2905-2916. doi: 10.21037/qims-24-1638. Epub 2025 Mar 28.
Current practice protocols for hypertrophic cardiomyopathy (HCM) recommend integrating crucial high-risk echocardiographic features associated with adverse prognosis into patient management. However, echocardiography is resource-limited and few powerful screening tools are available for risk assessment in the community. We aimed to devise an artificial intelligence (AI)-electrocardiogram (ECG) to identify high-risk echocardiographic features and monitor feature changes during long-term follow-up.
Patients with HCM from two hospitals who underwent both ECG and echocardiography within a 14-day window were retrospectively identified. One site (n=2,591) was used for training, validation and testing, and the other site (n=171) was used for external validation. An AI-ECG model was trained to predict the presence of any of the following four high-risk echocardiographic features: resting left ventricular outflow tract obstruction (LVOTO), massive left ventricular hypertrophy (LVH), systolic dysfunction, and apical aneurysm.
The AI-ECG model exhibited an area under the receiver operating characteristic curve (AUROC) of 0.81 [95% confidence interval (CI): 0.76-0.86] and 0.80 (95% CI: 0.72-0.87) in the testing and external validation set for identifying high-risk features. During a median follow-up of 37 months (interquartile range, 30-53 months), in the testing set, 30 patients with no high-risk features at baseline were evaluated. Ten patients developed new high-risk features, and changes in the predicted probability score was well matched to the trend of changes in risk status. Patients with false-positive results at baseline had a threefold increased risk of developing high-risk features compared with the true-negatives [hazard ratio (HR) 3.36, 95% CI: 0.77-14.65; P=0.037].
The AI-ECG model effectively identified and longitudinally monitored high-risk echocardiographic features and may serve as a powerful community screening tool.
目前肥厚型心肌病（HCM）的临床实践方案建议将与不良预后相关的关键高危超声心动图特征纳入患者管理。然而，超声心动图资源有限，社区中几乎没有强大的风险评估筛查工具。我们旨在设计一种人工智能（AI）心电图（ECG），以识别高危超声心动图特征并在长期随访期间监测特征变化。
回顾性确定两家医院在14天内同时接受心电图和超声心动图检查的HCM患者。一个研究点（n = 2591）用于训练、验证和测试，另一个研究点（n = 171）用于外部验证。训练了一个AI-ECG模型，以预测以下四种高危超声心动图特征中任何一种的存在：静息左心室流出道梗阻（LVOTO）、巨大左心室肥厚（LVH）、收缩功能障碍和心尖部室壁瘤。
在用于识别高危特征的测试集和外部验证集中，AI-ECG模型的受试者操作特征曲线下面积（AUROC）分别为0.81 [95%置信区间（CI）：0.76 - 0.86]和0.80（95% CI：0.72 - 0.87）。在中位随访37个月（四分位间距，30 - 53个月）期间，在测试集中，对30例基线时无高危特征的患者进行了评估。10例患者出现了新的高危特征，预测概率评分的变化与风险状态的变化趋势高度匹配。基线时假阳性结果的患者发生高危特征的风险是真阴性患者的三倍[风险比（HR）3.36，95% CI：0.77 - 14.65；P = 0.037]。
AI-ECG模型有效地识别并纵向监测了高危超声心动图特征，可作为一种强大的社区筛查工具。
