Clinical Report: Retinal AI Predicts Neonatal Lung Disease

Overview

Deep learning models utilizing retinal images from ROP screenings can predict bronchopulmonary dysplasia (BPD) and pulmonary hypertension (PH) in premature infants. The multimodal model combining image features and demographic data demonstrated superior diagnostic performance compared to models using either data type alone, specifically using a ResNet18 architecture for image feature extraction.

Background

Bronchopulmonary dysplasia is the most common chronic lung disease in premature infants and significantly contributes to morbidity and mortality, impacting long-term respiratory and neurodevelopmental outcomes. Early identification of infants at risk for BPD and PH is crucial for timely intervention and management. The integration of retinal imaging into routine care pathways presents a novel opportunity for risk assessment in this vulnerable population.

Data Highlights

Key Findings

• Multimodal models combining retinal images and demographic data outperformed single-modality models for predicting BPD.
• The imaging-only model achieved an AUC of 0.91 for predicting PH, indicating strong diagnostic capability.
• Retinal imaging may reveal systemic disease risk, supporting the emerging field of oculomics.
• Secondary models without visible ROP signs maintained consistent results, addressing confounding concerns.
• Findings suggest potential biological mechanisms linking retinal changes to cardiopulmonary conditions.
• Limitations include the small pulmonary hypertension cohort and lack of external validation across different imaging devices.

Clinical Implications

The ability to predict BPD and PH using retinal imaging could facilitate earlier interventions in high-risk infants. Incorporating this technology into existing screening protocols may enhance the management of premature infants and improve outcomes, though barriers such as training and resource allocation must be addressed.

Conclusion

This study highlights the potential of retinal AI as a predictive tool for neonatal lung disease, warranting further research and validation in clinical settings, including larger cohort studies and external validation.

References

1. JAMA Ophthalmology, 2026 -- Deep Learning–Based Prediction of Cardiopulmonary Disease in Retinal Images of Premature Infants
2. Ophthalmology Management, 2026 -- Study Shows Retinal Scans Predict Neonatal Lung Disease
3. Pulmonary hypertension in preterm neonates with bronchopulmonary dysplasia: a meta-analysis - PubMed
4. Ophthalmology Management — AI Advances for Diabetic Retinopathy
5. retinal physician — Study: AI Delivers High Accuracy in IRD Diagnosis
6. Retinal Physician — Study: AI Delivers High Accuracy in IRD Diagnosis
7. AI Advances for Diabetic Retinopathy
8. Study: AI Delivers High Accuracy in IRD Diagnosis
9. Pulmonary hypertension in preterm neonates with bronchopulmonary dysplasia: a meta-analysis - PubMed
10. Deep Learning–Based Prediction of Cardiopulmonary Disease in Retinal Images of Premature Infants | JAMA Ophthalmology | JAMA Network
11. Interventional Strategies for Children with Progressive Pulmonary Hypertension Despite Optimal Therapy: An Official American Thoracic Society Clinical Practice Guideline | American Journal of Respiratory and Critical Care Medicine