Singh Praveer, Kumar Sourav, Tyagi Riya, Young Benjamin K, Jordan Brian K, Scottoline Brian, Evers Patrick D, Ostmo Susan, Coyner Aaron S, Lin Wei-Chun, Gupta Aarushi, Erdogmus Deniz, Chan Rv Paul, McCourt Emily A, Barry James S, McEvoy Cindy T, Chiang Michael F, Campbell J Peter, Kalpathy-Cramer Jayashree
Ophthalmology, University of Colorado School of Medicine, Aurora, CO.
Radiology, MGH/Harvard Medical School, Charlestown, MA.
medRxiv. 2025 Sep 19:2025.09.18.25336004. doi: 10.1101/2025.09.18.25336004.
Bronchopulmonary dysplasia (BPD) and pulmonary hypertension (PH) are leading causes of morbidity and mortality in premature infants.
To determine whether images obtained as part of retinopathy of prematurity (ROP) screening might contain features associated with BPD and PH in infants, and whether a multimodal model integrating imaging features with demographic risk factors might outperform a model based on demographic risk alone.
A deep learning model was used to study retinal images collected from patients enrolled in the multi-institutional Imaging and Informatics in Retinopathy of Prematurity (i-ROP) study.
Seven neonatal intensive care units.
493 infants at risk for ROP undergoing routine ROP screening examinations from 2012 to 2020. Images were limited to <=34 weeks post-menstrual age (PMA) so as to precede the clinical diagnosis of BPD or PH.
BPD was diagnosed by the presence of an oxygen requirement at 36 weeks PMA, and PH was diagnosed by echocardiogram at 34 weeks. A support vector machine model was trained to predict BPD, or PH, diagnosis using: A) image features alone (extracted using Resnet18), B) demographics alone, C) image features concatenated with demographics. To reduce the possibility of confounding with ROP, secondary models were trained using only images without clinical signs of ROP.
For both BPD and PH, we report performance on a held-out testset (99 patients from the BPD cohort and 37 patients from the PH cohort), assessed by the area under receiver operating characteristic curve.
For BPD, the diagnostic accuracy of a multimodal model was 0.82 (95% CI: 0.72-0.90), compared to demographics 0.72 (0.60-0.82; P=0.07) or imaging 0.72 (0.61-0.82; P=0.002) alone. For PH, it was 0.91 (0.71-1.0) combined compared to 0.68 (0.43-0.9; P=0.04) for demographics and 0.91 (0.78-1.0; P=0.4) for imaging alone. These associations remained even when models were trained on the subset of images without any clinical signs of ROP.
Retinal images obtained during ROP screening can be used to predict the diagnosis of BPD and PH in preterm infants, which may lead to earlier diagnosis and avoid the need for invasive diagnostic testing in the future.
支气管肺发育不良（BPD）和肺动脉高压（PH）是早产儿发病和死亡的主要原因。
确定作为早产儿视网膜病变（ROP）筛查一部分所获得的图像是否可能包含与婴儿BPD和PH相关的特征，以及整合成像特征与人口统计学风险因素的多模态模型是否优于仅基于人口统计学风险的模型。
使用深度学习模型研究从参与多机构早产儿视网膜病变成像与信息学（i-ROP）研究的患者收集的视网膜图像。
七个新生儿重症监护病房。
2012年至2020年期间493名有ROP风险的婴儿接受常规ROP筛查检查。图像限于月经后年龄（PMA）<=34周，以便在BPD或PH的临床诊断之前。
BPD通过PMA 36周时的氧气需求诊断，PH通过34周时的超声心动图诊断。训练支持向量机模型以使用以下方法预测BPD或PH诊断：A）仅图像特征（使用Resnet18提取），B）仅人口统计学特征，C）与人口统计学特征连接的图像特征。为了降低与ROP混淆的可能性，仅使用没有ROP临床体征的图像训练二级模型。
对于BPD和PH，我们报告在一个保留测试集（来自BPD队列的99名患者和来自PH队列的37名患者）上的性能，通过受试者工作特征曲线下面积评估。
对于BPD，多模态模型的诊断准确性为0.82（95%CI：0.72-0.90），相比之下仅人口统计学特征为0.72（0.60-0.82；P=0.07）或仅成像为0.72（0.61-0.82；P=0.002）。对于PH，联合诊断准确性为0.91（0.71-1.0），相比之下人口统计学特征为0.68（0.43-0.9；P=0.04），仅成像为0.91（0.78-1.0；P=0.4）。即使在没有任何ROP临床体征的图像子集上训练模型，这些关联仍然存在。
ROP筛查期间获得的视网膜图像可用于预测早产儿的BPD和PH诊断，这可能导致早期诊断并避免未来进行侵入性诊断测试的需要。
