Detection of Jaundice in Neonates Using Artificial Intelligence

Abstract

Neonatal Jaundice is a condition that occurs in infants in the first few weeks of their birth. Jaundice tends to appear when the bilirubin level surpasses 5 mg/dl in the infants during their birth in nearly sixty percent of the full-term babies and eighty percent of the preterm babies. Techniques for automatic Jaundice detection have been developed, but the outcome is not accurate enough to attract the doctors for their usage. The aim of this project is for the early detection of Jaundice in neonates non-invasively and with greater accuracy. This study involved 37 normal infants and 22 Jaundice prone infants. The images of the infants need to be captured with a smartphone camera. Algorithms like face detection, skin detection, colour map transformation and white balancing were applied to the specific region of interest (ROI), and eight quantitative features were extracted from the processed image. Principle component analysis (PCA) was used in this study to minimize data redundancy. PCA indicated that features like skewness, entropy, standard deviation and mean possess more information for further classification of normal and Jaundice affected neonates. Extracted statistical features are fed as an input to the machine learners such as SVM and ensemble regression for further classification of infants as Jaundice affected or normal. The analysis result indicates that out of 37 neonates, 22 were classified as Jaundice affected babies, and 15 were normal. The machine learning approach showed an accuracy of 81.1%