人工智能通过单导联心电图检测左心室肥厚：可穿戴设备筛查的初步研究
Artificial intelligence-enabled detection of left ventricular hypertrophy using a single-lead electrocardiogram: a pilot study for wearable-based screening
左心室肥厚是心血管疾病和死亡率的独立预测因子。消费级可穿戴设备提供的单导联心电图，可能让无法做标准12导联心电图的场景也能筛查左心室肥厚。研究团队开发了一个基于Transformer架构的深度学习模型，用仅含第一导联的心电信号，并结合年龄、性别、体重指数等人口学变量，来检测左心室肥厚，并与同一框架下的12导联模型对比。数据来自PTB-XL数据集（21837份由心内科医生标注的心电图，来自18885名患者），分类阈值设定为使特异度达到90%以上，符合“筛查确诊”策略。结果显示，单导联模型AUC为0.874，加入人口学变量后升至0.928（灵敏度84.7%，特异度90.6%），已接近12导联加人口学变量的0.939。研究支持基于可穿戴设备筛查的可行性，但尚需与影像学确诊的左心室肥厚进行验证。
为什么推荐给您：单导联心电图AI筛查左心室肥厚的新应用，性能接近12导联，但为初步验证、未经影像确证。
不需要生物学背景，多打比方
正在获取全文并生成讲解（拿不到全文就依据摘要），大约需要 30–60 秒…
已等待 0 秒
这篇还没有动画
动画会把研究的流程、作用机制和关键结果一步一步演示出来，每一步都标明出自原文哪里。制作大约需要 30–60 秒。
摘要Abstract
BACKGROUND: Left ventricular hypertrophy (LVH) is an independent predictor of cardiovascular morbidity and mortality. Single-lead electrocardiogram (ECG) tracings from consumer wearable devices may enable scalable screening for LVH where standard 12-lead ECG acquisition is not feasible.
METHODS: We developed a transformer-based deep learning model to detect LVH using single-lead ECG (lead I) combined with demographic variables and benchmarked performance against a 12-lead ECG model within the same framework. The PTB-XL dataset (21,837 cardiologist-annotated ECGs from 18,885 patients) was used. Models were trained on (a) lead I only, (b) lead I plus demographics (age, sex, body mass index), and (c) 12-lead plus demographics. Classification thresholds were selected to achieve ≥90% specificity consistent with a rule-in screening paradigm.
RESULTS: The lead I-only model achieved an area under the curve (AUC) of 0.874 ± 0.062 (sensitivity 69.1% ± 18.9%, specificity 90.1% ± 0.1%). Incorporation of demographic variables improved performance to AUC 0.928 ± 0.017 (sensitivity 84.7% ± 4.2%, specificity 90.6% ± 0.7%). The 12-lead model with demographics achieved AUC 0.939 ± 0.007 (sensitivity 85.9% ± 2.0%, specificity 90.1% ± 0.1%).
CONCLUSIONS: Single-lead ECG combined with demographic variables enables detection of electrocardiographic LVH with performance approaching that of 12-lead ECG models, supporting the feasibility of wearable-based screening pending validation against imaging-confirmed LVH.