KEY POINTS:
- Researchers developed an artificial intelligence (AI) model that can analyze routine eye images of premature infants to detect risk for serious lung and heart diseases
- The study evaluated 493 premature infants across seven neonatal intensive care units (NICUs)
- AI predicted risk of bronchopulmonary dysplasia (BPD) with 82% accuracy and pulmonary hypertension (PH) with 91% accuracy
- Eye imaging is already standard care for screening retinopathy of prematurity (ROP), meaning this approach may not require additional procedures
- Earlier detection could help clinicians monitor vulnerable infants sooner and potentially reduce invasive diagnostic testing
Researchers have found that artificial intelligence (AI) can help identify serious lung and heart complications in premature infants by analyzing images already collected during routine eye screenings. These complications are major causes of illness and death in premature babies and often require invasive diagnostic testing.
In a new study published in JAMA Ophthalmology, researchers demonstrated that AI could detect signs of bronchopulmonary dysplasia (BPD) and pulmonary hypertension (PH) using retinal images captured to screen for retinopathy of prematurity (ROP). ROP is a potentially blinding eye disorder that commonly affects premature and low birthweight infants and requires screening in the first weeks of life.
“Artificial intelligence allows us to detect subtle patterns in retinal images that are not visible to the human eye,” said Praveer Singh, PhD, assistant professor of ophthalmology at the University of Colorado Anschutz and study lead author. “This opens the possibility of using a simple photograph to gain insights into a premature infant’s overall health.”
BPD is a chronic lung disease that frequently affects babies born very early, while PH is a dangerous form of high blood pressure that affects the lungs and heart. Both conditions can be difficult to diagnose early and often require specialized imaging or testing. Study findings suggest that AI analysis of retinal images could help clinicians identify these complications earlier and potentially reduce the need for more invasive testing.
“Premature and low birthweight babies undergo frequent eye imaging to screen for ROP,” said Jayashree Kalpathy-Cramer, PhD, professor of ophthalmology at CU Anschutz. “Our findings suggest that information about a baby’s lung and heart health may already be present in these images routinely collected in neonatal care. Earlier detection could make a meaningful difference in outcomes and treatment planning.”
Researchers analyzed retinal images collected during routine ROP screening exams from 493 infants receiving care at seven NICUs participating in a long-running, multi-institutional study supported by the National Institutes of Health. The study focused on images taken before clinical diagnoses of BPD or PH were typically made.
Using a deep learning model, investigators evaluated whether AI could detect disease-related patterns in the images. The model was tested using three approaches: imaging data alone, demographic and clinical risk factors alone, and a combined model incorporating both imaging and patient data.
The combined AI model demonstrated stronger diagnostic performance than demographic risk factors alone. For bronchopulmonary dysplasia, the combined model achieved 82% accuracy. For pulmonary hypertension, it achieved 91% accuracy.
Importantly, results remained consistent even when researchers excluded images showing clinical signs of ROP, suggesting the AI model identified information beyond traditional eye disease markers.
“One of the challenges with realizing the potential of many oculomics algorithms is that imaging the back of the eye is not (yet) part of the normal care pathway for many populations of patients,” says Peter Campbell, MD, MPH, professor of ophthalmology at Oregon Health and Science University and study co-author. “For more and more NICUs imaging IS part of the care pathway for ROP, which means the barriers to implement technologies like this are significantly lower"
Researchers say AI-assisted screening could eventually help clinicians identify vulnerable infants earlier and guide decisions about monitoring and treatment. While the findings are promising, investigators emphasize that further validation studies are needed before the technology could be integrated into routine clinical care.
About the Study
The research used data from infants enrolled in the multi-institutional Imaging and Informatics in Retinopathy of Prematurity (i-ROP) study, which has collected retinal imaging data from NICUs across the United States for more than a decade.
.png)
.png)
