Non-Invasive Neonatal Jaundice Detection Using Image Processing and Machine Learning

Neonatal hyperbilirubinemia, marked by elevated bilirubin levels, can lead to jaundice and severe complications if untreated. Conventional diagnosis involves invasive blood sampling, which is time-intensive and stressful for newborns. This study presents a non-invasive bilirubin estimation method using image processing, offering a faster, more accessible alternative. Key steps include skin detection, region of interest (ROI) extraction, and conversion to the YCbCr color space to enhance sensitivity to bilirubin-induced color shifts, especially in the Cb channel. Machine learning algorithms: K-Nearest Neighbors (k-NN), Random Forest (RF), and XGBoost, were evaluated, with XGBoost achieving 98% accuracy and the lowest mean squared error (MSE). This approach enables rapid jaundice detection, reducing reliance on repeated blood tests. Its integration into clinical practices and home-based monitoring systems offers the potential for early diagnosis and timely intervention, significantly improving neonatal healthcare outcomes.