Retinal AI Predicts Neonatal Lung Disease - Summary - MDSpire

Retinal AI Predicts Neonatal Lung Disease

  • March 16, 2026

  • 4 min

Share

Objective:

To evaluate whether retinal images from routine ROP screening can predict bronchopulmonary dysplasia and pulmonary hypertension in premature infants.

Key Findings:
  • The multimodal model for bronchopulmonary dysplasia achieved an AUC of 0.82, outperforming both demographics-only and imaging-only models (0.72 each).
  • The imaging-only model for pulmonary hypertension achieved an AUC of 0.91, significantly outperforming the demographics-only model (0.68).
  • Adding demographic data did not improve the performance of the multimodal model for pulmonary hypertension (AUC 0.91).
Interpretation:

The study suggests that retinal imaging can provide valuable predictive information for severe cardiopulmonary complications in premature infants, potentially enhancing clinical management.

Limitations:
  • Small pulmonary hypertension cohort limits statistical power.
  • Lack of external validation across different imaging devices.
  • Absence of model explainability analyses.
Conclusion:

Retinal imaging embedded in neonatal care pathways could support earlier identification of infants at high risk for severe cardiopulmonary complications, prompting timely interventions.

Original Source(s)

Related Content