AI Reads the Eyes to Predict Dangerous Pregnancy Complication Weeks in Advance
A new deep learning model analyzes retinal photographs to forecast preeclampsia risk with up to 97.9% accuracy, offering clinicians a noninvasive early warning system for maternal health.

Illustration: AI-generated
Preeclampsia, a potentially life-threatening pregnancy complication characterized by high blood pressure and organ damage, affects millions of expectant mothers worldwide and remains a leading cause of maternal and perinatal mortality. Now, researchers have demonstrated that artificial intelligence can predict this dangerous condition weeks before it develops—simply by examining photographs of the eye.
A multicenter study published in npj Digital Medicine describes Retina-PE, a multimodal deep learning model that combines retinal vascular imaging with standard clinical measurements to forecast preeclampsia risk. The approach achieved high accuracy in a prospective cohort of 8,616 pregnant individuals, suggesting that routine eye imaging could become a noninvasive screening tool in prenatal care.
Multimodal AI Achieves High Prediction Accuracy
The Retina-PE model integrates three data streams: retinal photographs capturing blood vessel patterns in the eye, maternal clinical factors, and mean arterial pressure measurements. Researchers from Shanghai First Maternity and Infant Hospital and Shanghai Jiading Maternal and Child Health Hospital developed the system using retrospectively analyzed data from prospectively collected multicenter cohorts.
In development datasets, the model demonstrated image-level AUCs ranging from 0.708 to 0.979 across various prediction timepoints. The system achieved an AUC of 0.979 for predicting early-onset preeclampsia one week in advance and 0.871 for identifying preterm preeclampsia within the same timeframe. For predicting preeclampsia occurrence at 24–28 weeks gestation, the model reached an AUC of 0.853.
External validation testing confirmed the model's robustness, with AUCs ranging from 0.728 to 0.815. The system maintained strong performance for predicting both preterm preeclampsia (AUC 0.801 for both 1-week and 4-week predictions) and early-onset disease (AUC 0.776 and 0.766 for 1-week and 4-week predictions, respectively).
Clinical Implications for Maternal Health
The retinal imaging approach addresses a gap in prenatal screening by offering a noninvasive method for early maternal risk stratification. Unlike invasive procedures or complex biomarker panels, retinal photography is quick, painless, and could potentially be integrated into routine prenatal visits.
The model's particularly strong performance in predicting early-onset preeclampsia—the most severe form of the condition—suggests potential for clinical deployment that could enable earlier interventions and more intensive monitoring for high-risk patients. This early warning system could help reduce emergency interventions and improve pregnancy outcomes by identifying at-risk mothers before symptoms become critical.
Research Support and Future Directions
The study was supported by the Science and Technology Commission of Shanghai Municipality, Shanghai Shen Kang Hospital Development Center, and Tongji University Medicine-X Interdisciplinary Research Initiative. Lead authors Zhou, T., Feng, W., Liu, H., and colleagues note that while the multicenter validation demonstrates promise, prospective clinical trials will be necessary to establish the model's real-world effectiveness and integration into standard prenatal care protocols.
Sources
Follow Health AI Central
Keep up with health AI where you already read
On LinkedIn
The day’s strongest health AI stories in your feed, one or two each weekday.
On Google
Choose Health AI Central as a preferred source, and Google is more likely to show our reporting in Top Stories.
Reporting & accountability
How we maintain accuracy
Health AI Central separates reported facts, analysis, and public-source data. Readers can review our standards, request a correction, or consult the relevant research methodology before relying on a claim.