AI Detects Illness in Infants

The new AI model specifically targets bronchopulmonary dysplasia (BPD), a chronic lung disease, and pulmonary hypertension (PH), a form of high blood pressure in the lungs. The research, published in *JAMA Ophthalmology*, involved 493 premature infants across seven neonatal intensive care units (NICUs). The AI demonstrated 82% accuracy in predicting BPD and 91% accuracy for PH by analyzing retinal images. This approach leverages a field known as "oculomics," which posits that the eye is a window to overall health. Subtle changes in the retina's blood vessels, not visible to the human eye, can indicate systemic issues. For instance, elevated pressure in the heart can affect retinal venous drainage, leaving a detectable signature in the eye's vasculature. These conditions are significant threats to premature infants. BPD is one of the most common and serious complications of premature birth. Pulmonary hypertension is a frequent complication of BPD, and the combination is particularly dangerous, with mortality rates as high as 48% within two years of a PH diagnosis. Current diagnostic methods for BPD and PH can be invasive and often come later in an infant's care. Diagnosis of BPD is typically confirmed around 36 weeks postmenstrual age if an infant still requires oxygen support. Suspected PH is usually investigated with an echocardiogram, but a definitive diagnosis might require cardiac catheterization, an invasive procedure. The AI analysis piggybacks on a routine and non-invasive procedure already performed on premature infants: screening for retinopathy of prematurity (ROP). This means vital information about lung and heart health could be gathered without adding new, stressful, or risky procedures for these vulnerable babies. The ability to identify at-risk infants weeks earlier than current methods could significantly alter outcomes. Early identification allows for closer monitoring and quicker intervention, potentially reducing the need for invasive tests and improving the prognosis for infants with these life-threatening conditions. This technology is part of a growing movement to use AI in neonatal care. A similar AI-based system for diagnosing ROP, the i-ROP DL system, has already received Breakthrough Device designation from the FDA. This suggests a potential pathway for the clinical integration of AI tools that can provide earlier, non-invasive diagnoses for the most fragile of patients.