Researchers at Seoul St Mary’s Hospital have created an AI-powered tool using MRI scans to identify structural brain markers linked to later speech and language delays in late-preterm babies, potentially enabling earlier intervention.
A team at Seoul St Mary’s Hospital has developed an artificial intelligence model that may help flag language-development problems in late-preterm babies well before standard speech tests can be done. In findings based on MRI scans taken around birth, the researchers found that a larger left amygdala and a smaller left hippocampus were associated with a higher risk of later language delay in children born between 34 and 37 weeks’ gestation, according to Newis and the underlying study indexed in PubMed.
The work centres on a difficult gap in paediatric care: many premature infants appear healthy after discharge, yet some later struggle with speech and language. Standard language assessment usually depends on direct testing by trained therapists, which is generally not practical until a child is old enough to cooperate, often after the age of three, the hospital said. By contrast, the new approach uses automated volumetry, a technique that measures brain regions from MRI scans and feeds those measurements into machine-learning software, to search for early structural clues.
Professor Kim Hyun-ho, who led the hospital’s work, said the model is not yet a diagnostic tool that can calculate an individual child’s exact odds of developing a disorder. Rather, he described it as a way to identify structural markers that appear statistically linked to later language delay. The hospital said the model was built from MRI data gathered in the newborn period and compared with language outcomes at ages two to six, with the aim of spotting children who may need closer follow-up sooner.
The findings fit with earlier research suggesting that premature birth can affect brain development in regions involved in emotion, memory and behaviour. Studies published in Frontiers in Human Neuroscience and The Journal of Pediatrics have reported altered amygdala development in very preterm infants and machine-learning models that can predict later language outcomes from early MRI features. The Seoul St Mary’s team says earlier identification could allow rehabilitation and developmental support to begin sooner, before language problems become more obvious.
Disclaimer: This content is for informational purposes only and is not intended to be a substitute for professional medical judgment, advice, diagnosis, or treatment.
