神经网络用于心电图诊断房颤的研究进展
Research Progress of Neural Network in Electrocardiographic Diagnosis of Atrial Fibrillation
DOI: 10.12677/ACM.2023.134939, PDF, HTML, XML, 下载: 257  浏览: 411  科研立项经费支持
作者: 石文海, 许 勇, 周 琳:成都市第六人民医院心内科,四川 成都;刘 婷:厦门大学经济学院统计系,福建 厦门;刘娟秀:电子科技大学光电学院,四川 成都;王吴婉:北京协和医院心脏超声科,北京;周 波:重庆医科大学附属第一医院心内科,重庆;黄 雄*:成都市第六人民医院普肝胆外科,四川 成都
关键词: 心房颤动心电图神经网络Atrial Fibrillation Electrocardiogram Neural Network
摘要: 房颤的发病具有一定的隐匿性,传统的临床实践对于部分房颤患者的早期识别存在不足;人工智能在心电图领域的应用越来越深入,神经网络模型可以对多种不同的心率失常进行识别和预测。本文就神经网络在心电图诊断房颤方面的新进展作一综述。
Abstract: Asmany episodes of atrial fibrillation remain asymptomatic, traditional clinical practice has defects in early identification of some patients with atrial fibrillation. Artificial intelligence is widely used in the field of electrocardiogram. Neural network models can identify and predict various kinds of ar-rhythmia. This article reviews the progress of neural network in electrocardiographic diagnosis for atrial fibrillation.
文章引用:石文海, 刘婷, 刘娟秀, 王吴婉, 周波, 许勇, 周琳, 黄雄. 神经网络用于心电图诊断房颤的研究进展[J]. 临床医学进展, 2023, 13(4): 6712-6721. https://doi.org/10.12677/ACM.2023.134939

1. 前言

自1628年William Harvey首次提出“房颤”的概念至今,房颤已成为全球性的卫生健康问题 [1] 。总体而言,房颤的患病率随年龄而增加,40岁左右人群患病率约1%,而超过70岁的人群比例达6.7% [2] 。这一比例有可能被低估,因为部分房颤患者的发病较为隐匿,没有明显的临床症状,尤其是阵发性房颤患者,常规体检心电图往往遗漏,直至出现栓塞并发症才得以诊断 [3] 。根据美国疾控中心数据,每年有15万余人因房颤或其并发症导致死亡;每年有45万余人因“房颤”反复住院 [4] 。因此,降低房颤病死率的关键在于早期筛查和诊断。近20年来随着神经网络在心电图领域的突破性进展,让房颤的早期诊断、并发症的预测逐步成为可能 [5] [6] 。

2. 神经网络对房颤心电图的自动识别

以往基于12导联心电图分析模型多是依赖传统的统计学算法,需要严格的特征提取和工程学计算,并且这种模型的误诊率并不低 [7] 。近期出现的“端–对–端”(end-to-end)深度神经网络模型(deep neural network, DNN)能自动分析心电图数据特征,省去了计算机提取特征的过程 [8] 。Hannun AY等首先通过便携式穿戴设备采集单一导联的心电图,然后利用深度神经网络模型在包含91,232个心电图的数据集上进行训练,最终识别出12种类型的心律失常,包含房颤、室性心动过速、房性心动过速、室性早搏等 [9] 。结果显示曲线下面积(area under the curve, AUC)等于0.97,并且将结果与6位心电专家相比,DNN的平均F1评分 = 0.837 (阳性预测值和敏感性的调和平均值),超过了心电专家的平均值0.780。该研究还显示DNN对于房颤的识别也很高(AUC = 0.96),但值得注意的是,这个AUC值是基于328例小样本房颤心电图数据得出的,虽然在外部验证集的选择上,选取了大数据集,但是测试数据并不是随机选取的,因此该模型在房颤识别的可靠性仍值得商榷 [9] 。

Ribeiro AH等 [10] 用200万份12导联心电图训练DNN模型,这个模型可以识别出6种异常心电图,但是由于测试集的异常心电图占比并不高(以房颤为例,只有13例进入到测试集中),其预测的准确性可能并不高。尽管如此,这个试验仍然是“端–对–端”分析理念的范例,并且强调了在DNN建模中扩大测试数据的意义 [11] 。

在2017年美国国家通用医学科学院和国家生物医学成像和生物工程院共同支持的生理信号库、计算心脏病学挑战数据集中 [12] ,75支参赛团队使用8528份共享的单导联心电图数据作为外部验证集,然后对房颤、正常心律、其他心律失常、噪音四种心电图进行识别,各模型结果进行“头对头”比较。胜出的算法有随机森林、卷积神经网络(convolutional neural network, CNN)以及自回归神经网络。但是训练集仍不足以体现复合算法的优势,因为这需要大量的参数和超参数调整。值得注意的是,虽然这个测试集包含了311例房颤患者的3658份心电图,但仅有27.3%被用于算法排序 [12] 。

Anthony H. Kashou等 [13] 基于72万患者的250万份12导联心电图研发了一种将“注意力机制”和DNN相结合的模型,该复合模型能将结局转换成66个可读取的诊断代码,并且可实现多标签的预测。该模型已在49万份心电图的测试集中显现其效能(AUC > 0.98)。最近,该研究团队还将该模型与传统的计算机预测软件、人工判读进行头对头比较。结果显示,模型的可靠判读率为91%,计算机软件为86%,最终心电图专家人工判读为94%。这表明该模型在一些疑难心电图的判读潜力上已逐步接近专家意见 [14] 。

3. 神经网络对房颤的预测模型研究

鉴于部分房颤发病的隐匿性,Attia ZI等 [15] 研发了一个独特的AI-ECG-AF (artificial intelligence-electrocardiogram-atrial fibrillation)模型,在没有任何额外信息的情况下,通过“正常窦性心律”的心电图预测潜在房颤患者的可能性。这个模型的基本原理是基于房颤患者心肌的病理结构异常,例如心肌肥大、纤维化、腔室扩大等引起的心脏电结构改变在心电图上产生的细微特征,这些特征是我们用人工方法无法观察到的 [16] 。研究团队分析了18万名房颤患者和60万名正常人心电图数据。患者被随机分为三组,70%用于训练数据集,10%用于内部验证集,20%用于测试集。房颤组在心电图确诊之前有1个月的窗口期,在这个时间段内,将心电图中房颤和左心房重构相关的潜在标签纳入其中。为优化模型效能,选取了8个导联(I、II,V1-V6),在13万份窦性心律心电图、3千个确诊房颤的数据集上测试该模型。模型的效力表现良好,AUC值波动在0.87~0.90 [15] 。

AI-ECG-AF模型可以用于健康人群的广泛筛查,这样能更有效地纳入可能患有房颤的人群,以进一步随访和诊断;高度疑诊房颤的患者需要更长时程的心电监测 [14] [17] 。此外,AI-ECG-AF模型可以作为预测房颤风险评分的补充,既往的评分多是基于“弗莱明汉房颤评分”或“CHARGE-AF评分” [18] [19] 。AI-ECG-AF模型作为未来房颤独立预测指标的实用性,已在梅奥诊所的一项研究中得到了验证 [20] 。在该研究中,CHARGE-AF评分和AI-ECG-AF模型都显示出相同的预测效能(C统计值均为0.69),将两者结合后,整体预测性能略有提高。模型输出值 > 0.5的志愿者,2年的累积发病率为21.5%,10年的累计发病率为52.2% [20] 。

Kashou AH等 [21] 基于心电图研发出一种类似于AI-ECG-AF模型的房颤预测算法。该模型在基线时使用窦性心律的心电图进行训练,并且重点分析房颤确诊前1~3年的心电图。研究团队使用英国生物数据库评估了该模型的性能,并显示出与CHARGE-AF风险评分相当的可比性。在CHARGE-AF风险评分中添加该模型后,模型的预测效能有所提高,这表明通过心电图预测房颤潜力巨大,并且CHARGE-AF包含的特征可能正是模型预测的重要节点。在临床应用中,这些模型可以早期筛查出常规体检人群中的隐匿性房颤患者,为早期干预房颤、预防严重并发症提供先机 [21] 。

4. 神经网络评估房颤负荷

目前,房颤分为阵发性、持续性或永久性,但治疗方案的拟定通常依据是否确诊房来评判是否启动抗凝,因而对于隐匿性房颤的患者,房颤负荷重,却没有得到及时的抗凝治疗 [22] 。近期已提出用定量的方法来评估房颤的负荷,旨在特定监测期内反映出患者房颤发作次数或时间比例 [23] 。房颤负荷对于卒中具有预测价值,可以指导预防性抗栓治疗 [24] 。然而,即使可穿戴和可植入器械可以实现更长时间的监测,房颤的负担也很难确定。此外,抗凝所需的最小房颤持续时间也没有明确的界定 [25] 。更容易忽略的是无症状患者房颤负荷的评估,传统的CHA2DS2-VASc评分对于高卒中风险、口服抗凝药物的患者受益较大,而对于低–中危患者关注不足,并且强调权衡栓塞与出血风险,但是二者往往没有明确的界限,多数情况下是参考医生的经验制定管理方案 [26] [27] 。

已有一些临床研究尝试解决这些问题。Shashikumar SP等 [28] 利用DNN模型对24小时动态心电图数据进行分析来评估房颤负荷,这是第一次运用神经网络和注意力模型对长时程心电图数据进行分析处理。Freedman B等通过采集植入式心电设备的数据建立了多种模型,包括逻辑回归、随机森林、CNN、嵌入CHA2DS2-VASc评分的复合模型。其中CNN模型的AUC值为0.6,敏感性0.57,特异性0.66,表现一般。随机森林、复合模型则现过拟合现象 [29] 。尽管这些研究初步证明了研发长期心电图监测数据模型的可行性,借助动态心电图能监测24~72 h内的心律变化;可植入电子设备监测时间更长 [30] [31] ,但临床应用尚需要进一步探索和验证,才能准确评估房颤负荷、指导抗凝治疗,起到进一步提高房颤患者预后的目的。

5. 神经网络与光电体积描记结合

光电体积描记(photoplethysmography, PPG)是利用光源和光电检测器检测的脉冲压力波形来测量数据,这项技术已成熟用于检测氧饱和度、心率和呼吸频率 [32] 。在PPG数据中,房颤表现为可变的脉冲间间隔和形态,DNN模型通过提取这些图像自分析来诊断房颤 [33] 。研发一个新的DNN模型过程较为繁琐,目前已有研究将大数据集的模型经过微调应用于小数据集,例如将12导联的心电图数据的DNN模型降阶应用到PGG来源(如移动穿戴设备)的单导联心电图,结果显示模型的效力仍然优异(AUC 0.97) [28] 。运动手环、智能手表就是运用这一原理来检测房颤。在“华为智能穿戴–心脏研究”中,总共有187,912名志愿者使用智能穿戴设备进行PPG分析,0.23%的人群收到“疑似房颤”预警信息,对这些人群进行规律随访,其中87%最终临床诊断为房颤 [34] 。在“苹果手表–心脏研究”中,总共419,297名患者,只有2161名(0.52%)收到了脉搏不规则的预警信息,进一步进行远程医疗就诊,患者邮寄得到心电图贴片,规律佩戴7天,在450名返回数据的患者中,只有34%的患者被诊断为房颤 [35] 。Fitbit心脏研究是正在进行中的大型前瞻性研究 [36] 。正如作者所指出的,与传统的逻辑回归、随机森林等算法,DNN模型表现出了优异的性能;但是,为了得出更准确的结论,需要在相同的测试数据集上进行头对头比较 [36] 。

PPG与DNN相结合为全人群房颤筛查提供了一种简单、无创的方法,但也突显出大规模筛查房颤的困难,需要考虑效益与风险问题 [37] 。目前的局限性主要在于:1) 数据的准确性和有效性不高;2) 缺乏统一的可穿戴设备临床诊断标准;3) 穿戴设备的购买成本偏高 [38] [39] 。其次,在低危人群中发送潜在心律失常的通知可能导致更高的假阳性率,增加不必要的焦虑和医疗费用;而在高危人群中进行PPG筛查也需谨慎判断结果,目前发现可穿戴设备采集的PPG数据中大约有40%是无效的,多数是干扰、噪音和伪影 [40] 。

6. 局限性

迄今为止,大多数DNN模型都没有经过大规模、严格的评估,它们的结果仍然局限于自己的数据集,需要对其泛化能力进行不断改进。其次,DNN固有的“黑盒子”特征让许多专业医疗工作者持保留态度 [41] ,因此,可解释的DNN模型是下一步需要研究的重点。

7. 总结

目前各种基于心电图的DNN模型层出不穷,这些模型在诊断房颤、预测并发症方面显示出巨大的临床潜力。在未来这种新的诊断方法可能成为临床常规工作的一部分。但是严格的、大样本的测试是新模型用于临床的前提,要经过反复的外部验证和临床试验以优化模型的泛化能力,这也有助于提升患者、临床医生对人工智能诊断的接受度。

基金项目

四川省中医药管理局专项课题(2020LC0019)。

NOTES

*通讯作者。

参考文献

[1] Bai, J., Lu, Y., Wang, H. and Zhao, J. (2022) How Synergy between Mechanistic and Statistical Models is Impacting Research in Atrial Fibrillation. Frontiers in Physiology, 13, Article 957604.
https://doi.org/10.3389/fphys.2022.957604
[2] Pitman, B.M., Chew, S.-H., Wong, C.X., et al. (2022) Prevalence and Risk Factors for Atrial Fibrillation in a Semi-Rural Sub-Saharan African Population: The Heart of Ethiopia: Focus on Atrial Fibrillation (TEFF-AF) Study. Heart Rhythm O2, 3, 839-846.
https://doi.org/10.1016/j.hroo.2022.09.008
[3] Machino, T., Aonuma, K., Maruo, K., et al. (2023) Randomized Crossover Trial of 2-Week Garment Electrocardiogram with Dry Textile Electrode to Reveal Instances of Post-Ablation Recurrence of Atrial Fibrillation Underdiagnosed during 24-Hour Holter Monitoring. PLOS ONE, 18, e0281818.
https://doi.org/10.1371/journal.pone.0281818
[4] Wang, Y.-C., Xu, X., Hajra, A., et al. (2022) Current Ad-vancement in Diagnosing Atrial Fibrillation by Utilizing Wearable Devices and Artificial Intelligence: A Review Study. Diagnostics, 12, Article No. 689.
https://doi.org/10.3390/diagnostics12030689
[5] Jin, Y., Qin, C., Liu, J., et al. (2020) A Novel Domain Adaptive Residual Network for Automatic Atrial Fibrillation Detection. Knowledge-Based Systems, 203, Article ID: 106122.
https://doi.org/10.1016/j.knosys.2020.106122
[6] Wang, J. (2020) A Deep Learning Approach for Atrial Fibrilla-tion Signals Classification Based on Convolutional and Modified Elman Neural Network. Future Generation Computer Systems, 102, 670-679.
https://doi.org/10.1016/j.future.2019.09.012
[7] Liu, Z., Zhang, Y., Chen, Y., Fan, X. and Dong, C. (2020) De-tection of Algorithmically Generated Domain Names Using the Recurrent Convolutional Neural Network with Spatial Pyramid Pooling. Entropy, 22, Article No. 1058.
https://doi.org/10.3390/e22091058
[8] Ullah, H., Bin Heyat, M.B., Akhtar, F., et al. (2022) An End-to-End Car-diac Arrhythmia Recognition Method with an Effective DenseNet Model on Imbalanced Datasets Using ECG Signal. Computational Intelligence and Neuroscience, 2022, Article ID: 9475162.
https://doi.org/10.1155/2022/9475162
[9] Hannun, A.Y., Rajpurkar, P., Haghpanahi, M., et al. (2019) Cardiolo-gist-Level Arrhythmia Detection and Classification in Ambulatory Electrocardiograms Using a Deep Neural Network. Nature Medicine, 25, 65-69.
https://doi.org/10.1038/s41591-018-0268-3
[10] Ribeiro, A.H., Ribeiro, M.H., Paixão, G.M.M., et al. (2020) Au-tomatic Diagnosis of the 12-Lead ECG Using a Deep Neural Network. Nature Communications, 11, Article No. 1760.
https://doi.org/10.1038/s41467-020-15432-4
[11] Lai, J., Chen, Y., Han, B., et al. (2019) A DenseNet-Based Di-agnosis Algorithm for Automated Diagnosis Using Clinical ECG Data. Journal of Southern Medical University, 39, 69-75.
[12] Kleyko, D., Osipov, E. and Wiklund, U. (2020) A Comprehensive Study of Complexity and Performance of Automatic Detection of Atrial Fibrillation: Classification of Long ECG Recordings Based on the PhysioNet Compu-ting in Cardiology Challenge 2017. Biomedical Physics & Engineering Express, 6, Article ID: 025010.
https://doi.org/10.1088/2057-1976/ab6e1e
[13] Kashou, A.H., Ko, W.-Y., Attia, Z.I., et al. (2020) A Comprehen-sive Artificial Intelligence-Enabled Electrocardiogram Interpretation Program. Cardiovascular Digital Health Journal, 1, 62-70.
https://doi.org/10.1016/j.cvdhj.2020.08.005
[14] Rabinstein, A.A., Yost, M.D., Faust, L., et al. (2021) Arti-ficial Intelligence-Enabled ECG to Identify Silent Atrial Fibrillation in Embolic Stroke of Unknown Source. Journal of Stroke and Cerebrovascular Diseases, 30, Article ID: 105998.
https://doi.org/10.1016/j.jstrokecerebrovasdis.2021.105998
[15] Attia, Z.I., Noseworthy, P.A., Lopez-Jimenez, F., et al. (2019) An Artificial Intelligence-Enabled ECG Algorithm for the Identification of Patients with Atrial Fibrillation during Sinus Rhythm: A Retrospective Analysis of Outcome Prediction. Lancet, 394, 861-867.
https://doi.org/10.1016/S0140-6736(19)31721-0
[16] Jansen, H.J., Bohne, L.J., Gillis, A.M. and Rose, R.A. (2020) A Trial Remodeling and Atrial Fibrillation in Acquired Forms of Cardiovascular Disease. Heart Rhythm O2, 1, 147-159.
https://doi.org/10.1016/j.hroo.2020.05.002
[17] Schwamm, L.H., Kamel, H., Granger, C.B., et al. (2023) Predictors of Atrial Fibrillation in Patients with Stroke Attributed to Large- or Small-Vessel Disease: A Prespecified Secondary Analysis of the STROKE AF Randomized Clinical Trial. JAMA Neurology, 80, 99-103.
https://doi.org/10.1001/jamaneurol.2022.4038
[18] Bisson, A., Lemrini, Y., El-Bouri, W., et al. (2022) Prediction of Incident Atrial Fibrillation in Post-Stroke Patients Using Machine Learning: A French Nationwide Study. Clinical Re-search in Cardiology.
https://doi.org/10.1007/s00392-022-02140-w
[19] Lin, J.Y., Larson, J., Schoenberg, J., et al. (2022) Serial 7-Day Electrocardiogram Patch Screening for AF in High-Risk Older Women by the CHARGE-AF Score. JACC: Clinical Electrophysiology, 8, 1523-1534.
https://doi.org/10.1016/j.jacep.2022.08.024
[20] Christopoulos, G., Graff-Radford, J., Lopez, C.L., et al. (2020) Artificial Intelligence-Electrocardiography to Predict Incident Atrial Fibrillation: A Population-Based Study. Circulation: Arrhythmia and Electrophysiology, 13, e009355.
https://doi.org/10.1161/CIRCEP.120.009355
[21] Khurshid, S., Friedman, S., Reeder, C., et al. (2022) ECG-Based Deep Learning and Clinical Risk Factors to Predict Atrial Fibrillation. Circulation, 145, 122-133.
https://doi.org/10.1161/CIRCULATIONAHA.121.057480
[22] Hindricks, G., Potpara, T., Dagres, N., et al. (2021) 2020 ESC Guidelines for the Diagnosis and Management of Atrial Fibrillation Developed in Collaboration with the European Association for Cardio-Thoracic Surgery (EACTS): The Task Force for the Diagnosis and Management of Atrial Fibrillation of the European Society of Cardiology (ESC) Developed with the Special Contribution of the European Heart Rhythm Association (EHRA) of the ESC. European Heart Journal, 42, 373-498.
https://doi.org/10.1093/eurheartj/ehaa612
[23] Monahan, K.H., Bunch, T.J., Mark, D.B., et al. (2022) Influence of Atrial Fibrillation Type on Outcomes of Ablation vs. Drug Therapy: Results from CABANA. EP Europace, 24, 1430-1440.
https://doi.org/10.1093/europace/euac055
[24] Peigh, G. and Passman, R.S. (2023) “Pill-in-Pocket” Anticoagulation for Stroke Prevention in Atrial Fibrillation. Journal of Cardiovascular Electrophysiology.
https://doi.org/10.1111/jce.15866
[25] Lucà, F., Giubilato, S., Di Fusco, S.A., et al. (2021) Anticoagulation in Atrial Fibrillation Cardioversion: What Is Crucial to Take into Account. Journal of Clinical Medicine, 10, Article No. 3212.
https://doi.org/10.3390/jcm10153212
[26] Tiver, K.D., Quah, J., Lahiri, A., Ganesan, A.N. and McGavigan, A.D. (2021) Atrial Fibrillation Burden: An Update—The Need for a CHA2DS2-VASc-Afburden Score. EP Europace, 23, 665-673.
https://doi.org/10.1093/europace/euaa287
[27] Perino, A.C., Fan, J., Askari, M., et al. (2019) Practice Variation in Anticoagulation Prescription and Outcomes after Device-Detected Atrial Fibrillation. Circulation, 139, 2502-2512.
https://doi.org/10.1161/CIRCULATIONAHA.118.038988
[28] Shashikumar, S.P., Shah, A.J., Clifforfd, G.D. and Nemati, S. (2018) Detection of Paroxysmal Atrial Fibrillation Using Attention-Based Bidirectional Recurrent Neural Networks. Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, London, 19-23 August 2018, 715-723.
https://doi.org/10.1145/3219819.3219912
[29] Kamel, H., Bartz, T.M., Longstreth Jr., W.T., et al. (2021) Cardiac Mechanics and Incident Ischemic Stroke: The Cardiovascular Health Study. Scientific Reports, 11, Article No. 17358.
https://doi.org/10.1038/s41598-021-96702-z
[30] Gladstone, D.J., Spring, M., Dorian, P., et al. (2014) Atrial Fi-brillation in Patients with Cryptogenic Stroke. New England Journal of Medicine, 370, 2467-2477.
https://doi.org/10.1056/NEJMoa1311376
[31] Diemberger, I., Biffi, M., Lorenzetti, S., et al. (2018) Predictors of Long-Term Survival Free from Relapses after Extraction of Infected CIED. EP Europace, 20, 1018-1027.
https://doi.org/10.1093/europace/eux121
[32] Sijerčić, A. and Tahirović, E. (2022) Photoplethysmography-Based Smart Devices for Detection of Atrial Fibrillation. Texas Heart Institute Journal, 49, e21756.
https://doi.org/10.14503/THIJ-21-7564
[33] Saarinen, H.J., Joutsen, A., Korpi, K., et al. (223) Wrist-Worn Device Combining PPG and ECG Can Be Reliably Used for Atrial Fibrillation Detection in an Outpatient Setting. Frontiers in Cardiovascular Medicine, 10, Article 1100127.
https://doi.org/10.3389/fcvm.2023.1100127
[34] Guo, Y., Wang, H., Zhang, H., et al. (2019) Mobile Photoplethysmographic Technology to Detect Atrial Fibrillation. Journal of the American College of Cardiology, 74, 2365-2375.
https://doi.org/10.1016/j.jacc.2019.08.019
[35] Perez, M.V., Mahaffey, K.W., Hedlin, H., et al. (2019) Large-Scale Assessment of a Smartwatch to Identify Atrial Fibrillation. New England Journal of Medicine, 381, 1909-1917.
https://doi.org/10.1056/NEJMoa1901183
[36] Lubitz, S.A., Faranesh, A.Z., Atlas, S.J., et al. (2021) Rationale and Design of a Large Population Study to Validate Software for the Assessment of Atrial Fibrillation from Data Acquired by a Consumer Tracker or Smartwatch: The Fitbit Heart Study. American Heart Journal, 238, 16-26.
https://doi.org/10.1016/j.ahj.2021.04.003
[37] 余超, 周伟, 王涛, 等. 可穿戴设备支持心房颤动人群筛查与管理研究进展[J]. 中国全科医学, 2023, 26(1): 113-117.
[38] Curry, S.J., Krist, A.H., Owens, D.K., et al. (2018) Screening for Atrial Fibrillation with Electrocardiography: US Preventive Services Task Force Recommendation Statement. JAMA, 320, 478-484.
https://doi.org/10.1001/jama.2018.10321
[39] Davidson, K.W., Barry, M.J., Mangione, C.M., et al. (2022) Screening for Atrial Fibrillation: US Preventive Services Task Force Recommendation Statement. JAMA, 327, 360-367.
https://doi.org/10.1001/jama.2021.23732
[40] Chikwetu, L., Miao, Y., Woldetensae, M.K., et al. (2023) Does Deidentification of Data from Wearable Devices Give Us a False Sense of Security? A Systematic Review. Lancet Digit Health, 5, E239-E247.
https://doi.org/10.1016/S2589-7500(22)00234-5
[41] Sehrawat, O., Kashou, A.H. and Noseworthy, P.A. (2022) Artificial Intelligence and Atrial Fibrillation. Journal of Cardiovascular Electrophysiology, 33, 1932-1943.
https://doi.org/10.1111/jce.15440