Computer Vision–Based Artificial Intelligence Tool for Direct Bilirubin Jaundice Measurement in Newborns

Background Conjugated hyperbilirubinemia, characterized by elevated levels of direct bilirubin (DB) may indicate underlying hepatobiliary disorders, such as biliary atresia, and warrants further investigation. Objectives The aim of this study was to accurately measure jaundice, related to DB levels, in newborn infants using an artificial intelligence (AI)–based computer vision tool. Method The computer vision tool used data processing, color transformations, and contrast enhancement techniques. Additionally, a convolutional neural network was created to predict DB levels. The study included 80 infants for training and validation and 17 infants for retesting. Five photographs were taken from the face, neck-chest, abdomen, extremities, and back after blood was drawn to measure DB levels. The photographs were captured using a professional camera under white light in the neonatal intensive care unit. Data analysis involved calculating the margin of error, percentage margin of error, and correlation statistics. Results The retest findings were analyzed to determine the margins of error. The study revealed a 5.24% discrepancy between the AI-based computer vision tool and the DB values obtained from laboratory blood tests. Furthermore, a positive correlation was observed between patient blood values and the mean calculated by the AI system. Discussion This study concluded that DB measurements, conducted under appropriate conditions, were accurately determined using AI with a good level of precision. Subsequent research on total bilirubin is recommended.