Developing and evaluating a machine learning model to enable automated alert function for neocam newborn eye imaging: a proof-of-concept study

A machine learning model will be built directly into a newborn eye imaging device to automatically flag potential sight-threatening conditions. Newborns currently undergo eye screening for conditions like cataracts and retinopathy of prematurity, but the images must be reviewed by trained specialists—a step that can delay diagnosis, especially in settings with limited access to expert clinicians. This project uses de-identified images from the DIvO study to develop and test an in-device algorithm that can analyse images in real time and trigger an alert when abnormalities are detected. If the proof-of-concept succeeds, the technology could be integrated into standard neonatal care. The immediate impact would be faster identification of eye problems, reducing the risk of permanent vision loss in infants. For hospitals without round-the-clock specialist cover, an automated alert function could transform a screening tool into a diagnostic safety net. The research is focused on technical feasibility and clinical validation; it does not yet address deployment at scale or integration into existing workflows.