Automated Calibration-Free Neonatal Jaundice Detection Using Machine Learning

Neonatal jaundice is a common clinical condition during the first week of life and requires early detection to reduce the risk of severe complications associated with elevated bilirubin levels. Conventional diagnostic approaches, including invasive blood sampling and device-based screening, may be costly, time-consuming, or unsuitable for continuous monitoring in resource-constrained settings. This paper presents an automated, non-invasive, and calibration-free framework for neonatal jaundice detection from smartphone images. The proposed framework first detects neonatal facial regions using YOLOv8 and then applies the Segment Anything Model (SAM) to obtain accurate facial skin segmentation. The segmented regions are enhanced using Contrast Limited Adaptive Histogram Equalization (CLAHE), followed by image augmentation to improve model robustness. Color-based features are extracted from RGB, YCbCr, HSV, and LAB color spaces and evaluated using six machine learning classifiers: Multi-Layer Perceptron (MLP), Decision Tree (DT), KNearest Neighbors (KNN), Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost).