Precise and Prompt Identification of Jaundice in Infants using AI

Jaundice in newborns is a common disorder. A failure to diagnose and treat Jaundice at the appropriate time might result in brain damage or death. A painful and time-consuming invasive blood test is one of the current diagnostic methods, and expensive transcutaneous bilirubinometers are used for non-invasive testing. Multiple attempts have been made to create non-invasive diagnostic tools using a smartphone camera since regular monitoring is essential. However, the current research uses either statistical or traditional machine learning techniques on skin or eye images. Using deep transfer learning technique in this work relies on composite images of the human body, including the eyes and skin. Moreover, in this study, the developed popular conventional machine learning techniques like Multi-Layer Perceptron, XG Boost, Decision Tree, and Random Forest are used to compare their effectiveness to that of the transfer learning model. According to the findings, traditional models performed best with eyes and merged features, whereas the transfer learning model worked best with skin images. With an AI model trained on 760 images and tested on 670 images, the output indicates a likelihood of 98.693% that jaundice is present in the infant.