Sharma Puneet, Gearhart Addison, Luo Guangze, Palepu Anil, Wang Cindy, Mayourian Joshua, Beam Kristyn, Spyropoulos Fotios, Powell Andrew J, Levy Philip, Beam Andrew
Division of Neonatology, Emory University School of Medicine, Atlanta, Georgia.
Department of Cardiology, Boston Children's Hospital, Boston, Massachusetts.
J Am Soc Echocardiogr. 2025 Jul;38(7):624-632. doi: 10.1016/j.echo.2025.03.018. Epub 2025 Apr 11.
Patent ductus arteriosus (PDA) is associated with significant morbidity and mortality in preterm infants. Although pharmacotherapy can be effective, it is difficult to predict whether a patient will respond, leading to delays in care. Machine learning has emerged as a powerful tool to interpret clinical data to predict clinical outcomes but has not yet been applied to this question. The aim of this study was to train and validate a novel deep learning model to predict the likelihood of PDA closure after an initial course of pharmacotherapy in preterm infants.
A retrospective cohort of 174 preterm infants who received pharmacologic treatment for PDA was identified. After collecting relevant perinatal data and pretreatment echocardiograms, the subjects were randomized into training and validation sets in a 70:30 split. Two distinct convolutional neural networks (CNN) were trained, one based on echocardiograms alone and the other on both echocardiograms and perinatal data. The performance of the CNNs was compared against controls of random forest and logistic regression models trained on perinatal data alone.
The rate of PDA closure after an initial course of pharmacotherapy was 60% in this cohort. The 174 echocardiograms collected for all subjects included 1,926 clips. A total of 121 infants (1,387 clips) were successfully randomized into the training set and 53 (539 clips) into the validation set. The multimodal CNN had an area under the curve (AUC) of 0.82, outperforming the imaging-only model (AUC = 0.66). Additionally, the multimodal CNN outperformed logistic regression (AUC = 0.66) and random forest (AUC = 0.74) models.
This novel, multimodal CNN shows promise for clinicians, who do not currently have a reliable tool to predict the success of PDA closure after an initial course of pharmacotherapy. This investigation represents the first attempt to use deep learning methodology to predict this outcome.
动脉导管未闭（PDA）与早产儿的显著发病率和死亡率相关。尽管药物治疗可能有效，但难以预测患者是否会有反应，从而导致治疗延迟。机器学习已成为解释临床数据以预测临床结果的强大工具，但尚未应用于这个问题。本研究的目的是训练并验证一种新型深度学习模型，以预测早产儿在初始药物治疗疗程后动脉导管未闭闭合的可能性。
确定了一个接受PDA药物治疗的174例早产儿的回顾性队列。收集相关围产期数据和治疗前超声心动图后，将受试者按70:30的比例随机分为训练集和验证集。训练了两个不同的卷积神经网络（CNN），一个仅基于超声心动图，另一个基于超声心动图和围产期数据。将CNN的性能与仅基于围产期数据训练的随机森林和逻辑回归模型的对照组进行比较。
该队列中，初始药物治疗疗程后PDA闭合率为60%。为所有受试者收集的174份超声心动图包括1926个片段。共有121名婴儿（1387个片段）成功随机分配到训练集，53名（539个片段）分配到验证集。多模态CNN的曲线下面积（AUC）为0.82，优于仅基于影像的模型（AUC = 0.66）。此外，多模态CNN优于逻辑回归（AUC = 0.66）和随机森林（AUC = 0.74）模型。
这种新型的多模态CNN对临床医生来说很有前景，因为他们目前没有可靠的工具来预测初始药物治疗疗程后PDA闭合的成功率。这项研究是首次尝试使用深度学习方法来预测这一结果。
