AI Scans Infant Eyes to Detect Illness
Researchers have developed an AI model that can analyze routine eye images of premature infants to detect their risk for serious lung and heart conditions. The noninvasive technique could help identify life-threatening complications much earlier in preterm babies.
The new AI model detects bronchopulmonary dysplasia (BPD), a chronic lung disease, and pulmonary hypertension (PH), a type of high blood pressure affecting the lungs and heart. The AI predicted BPD with 82% accuracy and PH with 91% accuracy in a study of 493 premature infants across seven neonatal intensive care units (NICUs). This "oculomics" approach — using the eye to understand systemic disease — analyzes subtle patterns in retinal blood vessels that are not visible to the human eye. Researchers suggest that oxygen therapy and mechanical ventilation used to treat premature infants may alter the retinal vasculature in ways the AI can detect. Similarly, elevated pressure from pulmonary hypertension could affect retinal blood flow, leaving a detectable signature. Current diagnosis for BPD often relies on observing a prolonged need for oxygen support and can involve chest X-rays or CT scans. For pulmonary hypertension, the standard non-invasive screening tool is an echocardiogram, but a definitive diagnosis may require an invasive cardiac catheterization. The AI analysis uses images already being captured to screen for an eye disease common in premature infants called retinopathy of prematurity (ROP). This means the new diagnostic insight could be gained without requiring any additional, invasive procedures for these fragile babies. The research, led by Praveer Singh, PhD, at the University of Colorado Anschutz, is currently a "proof of concept." Next steps involve validating the AI's effectiveness with different types of imaging equipment and developing a clinical workflow to integrate the tool into NICU care.
