Deep learning retinal vascular biomarker predicts preeclampsia in a multicenter prospective cohort

Preeclampsia (PE) causes maternal and perinatal morbidity and mortality, yet noninvasive screening remains limited. We developed Retina-PE, a multimodal deep learning model integrating retinal photographs, maternal clinical factors, and mean arterial pressure to predict subsequent PE. Prospectively collected multicenter data from 8616 pregnant individuals were used for retrospective model development and evaluation. Development-dataset image-level areas under the receiver operating characteristic curve (AUCs) were 0.853 and 0.708 for PE occurrence at 24–28 and 28–32 weeks, 0.871 and 0.791 for 1- and 4-week preterm PE, and 0.979 and 0.862 for 1- and 4-week early-onset PE. Corresponding external-validation-dataset AUCs were 0.728 and 0.815, 0.801 and 0.801, and 0.776 and 0.766. Incremental performance was task-specific. Patient-level analyses showed similar patterns but included few events in some tasks. These findings support prospective evaluation of retinal imaging for maternal risk stratification.

Authors

Institutions

Publication Details

Journal
npj Digital Medicine
Published
2026-11-17
DOI
https://doi.org/10.1038/s41746-026-03463-y
Primary Topic
Pregnancy and preeclampsia studies
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Deep learning retinal vascular biomarker predicts preeclampsia in a multicenter prospective cohort

Zongyuan Ge, Huanrong Liu, Jianhao Xiong, Shengyi Gu et al.
2 links
npj Digital Medicine
Pregnancy and preeclampsia studies
article

Deep learning retinal vascular biomarker predicts preeclampsia in a multicenter prospective cohort

Zongyuan Ge, Huanrong Liu, Jianhao Xiong, Shengyi Gu, Tianfan Zhou, Xiaolin Hua, Jing Peng, Huan Wang, Sheng Biao Wan, Weijia Luo, Yajing Mao, Ping Li, Wei Feng, Peng Gao, Qin Lu, Rongqing Zhang
article en

Abstract

Preeclampsia (PE) causes maternal and perinatal morbidity and mortality, yet noninvasive screening remains limited. We developed Retina-PE, a multimodal deep learning model integrating retinal photographs, maternal clinical factors, and mean arterial pressure to predict subsequent PE. Prospectively collected multicenter data from 8616 pregnant individuals were used for retrospective model development and evaluation. Development-dataset image-level areas under the receiver operating characteristic curve (AUCs) were 0.853 and 0.708 for PE occurrence at 24–28 and 28–32 weeks, 0.871 and 0.791 for 1- and 4-week preterm PE, and 0.979 and 0.862 for 1- and 4-week early-onset PE. Corresponding external-validation-dataset AUCs were 0.728 and 0.815, 0.801 and 0.801, and 0.776 and 0.766. Incremental performance was task-specific. Patient-level analyses showed similar patterns but included few events in some tasks. These findings support prospective evaluation of retinal imaging for maternal risk stratification.

npj Digital Medicine
Tongji University (CN), Shanghai Jiao Tong University (CN), Jiading District Central Hospital (CN), Pudong Maternal and Child Health Hospital (CN), International Peace Maternity & Child Health Hospital (CN), Shanghai Tenth People's Hospital (CN), Shanghai First Maternity and Infant Hospital (CN), Monash University (AU)
Science and Technology Commission of Shanghai Municipality, Shanghai Shen Kang Hospital Development Center
Good health and well-being
Pregnancy and preeclampsia studies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon