New AI model uses retinal images to help predict PH risk in infants
Study: Earlier detection could influence outcomes, treatment planning
Written by |
Scientists have developed an artificial intelligence-powered model that may help identify premature infants at risk for pulmonary hypertension (PH) and bronchopulmonary dysplasia (BPD) using noninvasive eye pictures taken during standard screenings.
“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,” Jayashree Kalpathy-Cramer, PhD, professor of ophthalmology at the University of Colorado Anschutz, and the study’s senior author, said in a university press release.
The study, “Deep Learning–Based Prediction of Cardiopulmonary Disease in Retinal Images of Premature Infants,” was published in the journal JAMA Ophthalmology.
PH can be complication of BPD and increase risk of serious disease
BPD is the most frequent chronic lung disease in premature infants. It causes injury to the airways (bronchi) and interferes with the development of the tiny air sacs in the lungs. PH can be a complication of BPD and increases the risk of serious disease.
Diagnosing BPD and PH early in preterm babies can be challenging, since BPD prediction tools and PH diagnosis rely on tests and techniques that may not be sensitive enough or may be resource- and time-intensive.
“Obtaining alternative biomarkers correlated with earlier stages of BPD and PH through noninvasive techniques could avoid the need for more invasive testing and potentially help neonatologists … enabling more individually tailored care,” the investigators wrote.
With this in mind, the researchers developed an AI-powered model to analyze retinal images taken during routine screenings for retinopathy of prematurity (ROP). ROP affects the retina — the light-sensing tissue at the back of the eye — in premature babies.
A total of 493 premature infants were included in the analysis, which used retinal images collected between June 2015 and April 2020 as part of the Imaging and Informatics in Retinopathy of Prematurity (i-ROP) study across the U.S.
“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,” said Peter Campbell, MD, professor of ophthalmology at Oregon Health and Science University and a study co-author. “For more and more NICUs [neonatal intensive care units], imaging IS part of the care pathway for ROP, which means the barriers to implement technologies like this are significantly lower.”
Combined model performed better
The images were taken at 34 weeks or less of postmenstrual age (the time from the first day of the mother’s last menstrual period), so they were captured before BPD and PH are typically diagnosed.
To assess whether eye photos could signal emerging lung and heart problems, the researchers trained an AI model using three approaches: retinal imaging alone, demographic and clinical risk factors alone, and a combined model integrating both.
In the study, BPD was defined as a baby needing supplemental oxygen at 36 weeks of postmenstrual age, while PH was defined based on an echocardiogram at 34 weeks’ postmenstrual age and certain pulmonary artery pressure criteria.
The team then tested how well the AI model could predict which infants would meet those criteria using a standard measure called the area under the receiver operating characteristic (ROC) curve (AUC). AUC values range from 0 to 1, with 0.5 indicating random chance and higher values indicating better performance at distinguishing between babies with the condition and those without it.
Retinal images obtained during ROP screening may be used to predict the diagnosis of BPD and PH in preterm infants, which may lead to earlier diagnosis and avoid the need for invasive diagnostic testing in the future.
Overall, the combined model performed better. For PH, the combined model achieved an AUC of 0.91, the same as imaging only, but better than 0.68 for demographic risk factors alone.
For BPD, the combined model reached an AUC of 0.82, compared with 0.72 for both demographics-only and imaging-only models.
Importantly, the results held up when the researchers removed images showing signs of ROP, suggesting the model wasn’t simply “reading” eye disease severity.
“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 the study’s lead author. “This opens the possibility of using a simple photograph to gain insights into a premature infant’s overall health.”
The researchers stressed the model’s potential in aiding diagnosis.
“Overall, these findings suggest that “retinal images obtained during ROP screening may be used to predict the diagnosis of BPD and PH in preterm infants, which may lead to earlier diagnosis and avoid the need for invasive diagnostic testing in the future,” they concluded.
