Non-invasive and non-contact automatic jaundice detection of infants based on random forest

Jaundice is a common phenomenon in neonates and a significant cause of morbidity and mortality. Early detection of jaundice is still a significant challenge. Moreover, existing invasive techniques are stressful, and painful for the newborn, and non-invasive devices are expensive. Therefore, we investigate the characteristics of a non-invasive and non-contact neonatal jaundice detection system based on skin colour analysis and machine learning using a graphical user interface (GUI). First, we automatically selected a region of interest (ROI) from the image of an infant captured by a digital camera. Then, the skin colour of the selected ROI was analysed in both RGB and YCbCr colour spaces. Finally, a machine learning algorithm based on random forest (RF) was incorporated to classify jaundice or normal and determine whether the neonate requires treatment or not. The experimental result demonstrates that the proposed jaundice detection system has the potential as a non-invasive technique in clinical application.