Real Time Diagnosis of Neonatal Jaundice Using Machine Learning

Abstract

Neonatal jaundice, a condition commonly found among newborns, usually requires invasive and timely methods for diagnosis to prevent further complications. This research presents a real-time diagnostic system for neonatal jaundice using machine learning and image processing techniques. The system utilizes a dataset of neonatal images, which undergo preprocessing to extract relevant features. Features, including color values from different color spaces, are analyzed using multiple machine learning models, such as XGBoost, CatBoost, Support Vector Machines (SVM), Random Forest (RF), and LightGBM. These models are trained and evaluated for their predictive performance. A user-friendly graphical user interface is developed to enable real-time diagnosis, implemented on a Raspberry Pi device equipped with a webcam to acquire real-time image capture and apply image processing. The system demonstrates the potential for accessible and reliable neonatal care solutions.