An artificial intelligence algorithm has demonstrated the ability to predict bronchopulmonary dysplasia (BPD) and pulmonary hypertension (PH) in preterm infants by analysing retinal images taken during routine eye screening. The findings, published online in JAMA Ophthalmology, highlight the emerging potential of retinal-based diagnostics, known as “oculomics”, in neonatal risk assessment.
Researchers analysed data from 493 infants at risk for retinopathy of prematurity (ROP) across seven US neonatal intensive care units. The team trained deep learning models using three approaches: retinal image features alone, demographic variables alone (including birth weight and gestational age), and a multimodal combination of both.
For BPD prediction, the multimodal model achieved higher diagnostic accuracy than either demographics-only or image-only models. For PH prediction, both the image-only and multimodal models performed substantially better than demographics alone, suggesting that retinal features may independently identify PH risk.
The researchers noted that retinal images appeared to complement demographic risk factors for BPD, whilst providing strong independent diagnostic value for PH. Secondary analysis confirmed these findings were not confounded by underlying ROP diagnosis.
Corresponding author J. Peter Campbell of Oregon Health & Science University and colleagues emphasised that oculomics has shown promise for diagnosing cardiovascular, neurodegenerative, and kidney disorders through retinal imaging. However, the authors described these findings as hypothesis-generating and stressed the need for validation in larger, more diverse populations before clinical implementation.
The study suggests that retinal imaging during routine ROP screening may offer a non-invasive approach to early risk stratification for preterm infants.
Source: Physicians Weekly / Oregon Health & Science University (JAMA Ophthalmology, 2026)
