基于超声心动图的深度学习慢性肾脏病筛查
Deep learning-enabled screening of chronic kidney disease from echocardiography.
一句话超声心动图深度学习模型可识别慢性肾脏病,在两个独立外部队列中保持稳健表现。
研究问题
数据与样本
分析流程
- 1模型开发:使用CSMC患者的PLAX视频训练检测CKD的深度学习模型。
- 2内部测试:在CSMC留出测试队列中评估任意分期CKD的检测AUC。
- 3外部验证:在SHC和KPNC两个独立临床队列中评估模型AUC。
- 4亚组评估:分别考察伴有和不伴有代谢性合并症患者中的模型表现。
关键方法与工具
- deep learning (DL)
- 从超声视频识别CKD
- parasternal long-axis (PLAX)
- 提供超声心动图视频输入
- area under the curve (AUC)
- 评价模型区分能力
设计思路
主要发现
- CSMC留出测试队列中,模型检测任意分期CKD的AUC为0.756(95%置信区间:0.749-0.763)。
- 外部验证中,KPNC队列AUC为0.718 [0.714-0.723],SHC队列AUC为0.719 [0.704-0.735]。
- 亚组结果:伴有及不伴有代谢性合并症的患者中,模型均表现良好,提示其可能学习到了CKD特异性的影像特征。
可借鉴之处
- 开发影像筛查模型时,可借鉴本文将中心内留出测试与两个独立机构外部验证结合的路径,检验跨机构适用性。
- 可借鉴本文按有无代谢性合并症进行亚组评估的做法,考察模型表现是否依赖相关合并症。
数据与代码
摘要(中文翻译)
背景:慢性肾脏病(CKD)影响全球近850 million人。未被诊断的CKD比例超过60%。
方法:利用CKD与心血管疾病之间的关系,我们使用Cedars-Sinai Medical Center(CSMC)62,818例患者的325,377段胸骨旁左心室长轴(PLAX)视频,开发了从PLAX视频检测CKD的深度学习(DL)模型。我们在Stanford Healthcare(SHC)的2,762例患者和Kaiser-Permanente Northern California(KPNC)的41,611例患者这2个独立队列中进行了模型外部验证。
结果:在CSMC的留出测试队列中,模型检测任意分期CKD的曲线下面积(AUC)为0.756(95%置信区间:0.749-0.763),并在KPNC(AUC:0.718 [0.714-0.723])和SHC(AUC:0.719 [0.704-0.735])队列中持续表现良好。模型在伴有及不伴有代谢性合并症的亚组中均表现良好,提示其学习到了CKD特异性的影像特征。
结论:我们的DL超声心动图模型在2个外部临床中心均以稳健的性能检测出CKD,从而为无创筛查和提高检出率提供了一条途径。
资助:本研究获得Sarnoff Cardiovascular Research Foundation、American Heart Association(25POST1357984和25AHAI1487693)以及National Institutes of Health(R00HL157421、R01HL173487和R01HL173526)的支持。
英文原文摘要
Background:Chronic kidney disease (CKD) affects nearly 850 million individuals globally. The prevalence of undiagnosed CKD is over 60%.
Methods:Taking advantage of the relationship between CKD and cardiovascular disease, we developed a deep learning (DL) model to detect CKD from parasternal long-axis (PLAX) videos, using 325,377 PLAX videos from 62,818 patients at Cedars-Sinai Medical Center (CSMC). We externally validated our model in two independent cohorts of 2,762 patients at Stanford Healthcare (SHC) and 41,611 patients at Kaiser-Permanente Northern California (KPNC).
Findings:In a held-out test cohort at CSMC, our model detected any stage of CKD with an area under the curve (AUC) of 0.756 (95% confidence interval: 0.749-0.763), with consistently strong performance in the KPNC (AUC: 0.718 [0.714-0.723]) and SHC (AUC: 0.719 [0.704-0.735]) cohorts. Our model performed well across subgroups with and without metabolic comorbidities, suggesting that it learned imaging features specific to CKD.
Conclusions:Our DL echo model detected CKD with robust performance at two external clinical sites, thus offering an avenue for noninvasive screening and improved detection rates.
Funding:This work was supported by the Sarnoff Cardiovascular Research Foundation, the American Heart Association (25POST1357984 and 25AHAI1487693), and the National Institutes of Health (R00HL157421, R01HL173487, and R01HL173526).
摘要来自 PubMed。翻译与解读由模型根据摘要生成,可能有误,不构成诊疗建议;数据以原文为准,版权归原出版方所有。