AI Predicts Infant Pulmonary Hypertension Risk with Retinal Scan

March 7, 2026 Nkechi Okonkwo- Health Editor Health

A new artificial intelligence model shows promise in identifying premature infants at risk for pulmonary hypertension (PH) and bronchopulmonary dysplasia (BPD) using standard retinal images taken during routine screenings for retinopathy of prematurity (ROP). The findings, published in JAMA Ophthalmology, suggest that crucial information about a baby’s lung and heart health may be visible in these routinely collected eye images, potentially leading to earlier detection and improved treatment planning.

Understanding Pulmonary Hypertension and Bronchopulmonary Dysplasia

Pulmonary hypertension is a condition where there is high blood pressure in the arteries that go from the heart to the lungs. In infants, this can lead to serious complications. Bronchopulmonary dysplasia (BPD), the most frequent chronic lung disease in premature infants, causes injury to the airways and interferes with the development of the tiny air sacs in the lungs. PH can develop as a complication of BPD, further increasing the risk of severe illness. Diagnosing both conditions early in preterm babies can be challenging, as current diagnostic tools can be resource-intensive and may not always be sensitive enough.

The research team, led by Jayashree Kalpathy-Cramer, PhD, professor of ophthalmology at the University of Colorado Anschutz, developed the AI model to address this challenge. The model analyzes retinal images – pictures of the back of the eye – to identify patterns associated with the development of PH and BPD. This approach is particularly appealing because retinal screenings are already a standard part of care for premature infants, making it a non-invasive and potentially cost-effective way to identify at-risk babies.

How the AI Model Works

Researchers analyzed retinal images from 493 infants at risk for ROP across seven US neonatal intensive care units. The infants had a postmenstrual age of 34 weeks or less at the time of screening. Clinicians diagnosed BPD based on oxygen requirements at 36 weeks’ postmenstrual age and PH using echocardiography at 34 weeks’ postmenstrual age. The team then trained different deep learning models to predict BPD or PH. These models used three approaches: analyzing retinal image features alone, using demographic variables (postmenstrual age, birth weight, and gestational age) alone, and combining both types of data.

The study found that combining retinal image features with demographic variables improved the accuracy of BPD diagnosis. Importantly, the image-based models alone were strongly able to identify PH, highlighting the potential of this “oculomics” approach – using retinal imaging to diagnose non-eye diseases – in neonatal care. Physicians Weekly reports that the research demonstrates the ability of deep learning algorithms to diagnose non-ocular diseases from retinal images.

The Promise of ‘Oculomics’

This research builds on a growing field known as oculomics, which explores the potential of using retinal images to diagnose a wide range of conditions beyond eye diseases. Recent advances in deep learning have shown that algorithms can identify signs of cardiovascular disease, neurodegenerative disorders, kidney problems, and neurological conditions in retinal images. The eye, with its direct connection to the brain and extensive network of blood vessels, offers a unique window into overall health.

J. Peter Campbell, MD, MPH, of Oregon Health & Science University, and colleagues, writing in JAMA Ophthalmology, noted that oculomics has shown promise for the retinal-based diagnosis of numerous diseases. This study adds to the evidence supporting the idea that retinal imaging can be a valuable tool for early disease detection and risk stratification.

What So for Premature Infants

Early detection of PH and BPD is crucial because it allows clinicians to intervene sooner and potentially improve outcomes. Current diagnostic methods, such as echocardiography for PH, can be time-consuming and require specialized expertise. An AI-powered model that can quickly and accurately identify at-risk infants could streamline the diagnostic process and ensure that babies receive the care they need in a timely manner.

However, it’s important to note that this research is still in its early stages. The AI model needs to be further validated in larger and more diverse populations. The study’s findings suggest a potential for improved screening, but do not replace the need for clinical evaluation and established diagnostic procedures.

Looking Ahead: Further Research and Clinical Implementation

The researchers emphasize that further research is needed to refine the AI model and assess its clinical utility. Future studies will likely focus on:

  • Expanding the dataset to include more infants from different geographic locations and with varying levels of prematurity.
  • Investigating the specific retinal features that are most predictive of PH and BPD.
  • Developing a user-friendly interface that allows clinicians to easily access and interpret the AI model’s results.
  • Conducting clinical trials to evaluate the impact of the AI model on patient outcomes.

The development of this AI model represents a significant step forward in the effort to improve the care of premature infants. While more research is needed, the potential for earlier detection and more effective treatment of PH and BPD is promising. The Pulmonary Hypertension News article highlights that earlier detection could build a meaningful difference in outcomes and treatment planning.