JAUNDICE, marked by yellowing of the skin and eyes due to elevated bilirubin levels, can signal various underlying health conditions such as liver disease, bile duct obstruction, or haemolysis. Early detection and accurate diagnosis are essential for timely intervention, particularly in resource-limited settings. Traditionally, bilirubin is measured through blood tests, but no dedicated, non-invasive screening method for jaundice exists.
A recent study has demonstrated the potential of using AI to predict jaundice by analysing scleral images and urine colour. Leveraging advancements in machine learning and image processing, researchers developed an AI program capable of identifying signs of jaundice with high accuracy. This method offers a fast, non-invasive alternative to traditional blood testing.
The programme utilises deep learning algorithms, including DeepSets and Random Forest models, to analyse images of the sclera – the white part of the eye – taken against an A4 white paper for colour correction. The study found that this technique achieved an accuracy of 87.1% in distinguishing jaundiced from non-jaundiced individuals, with a strong correlation (0.79) between predicted and actual bilirubin levels. These results are comparable to, and in some cases rival, previous research such as BiliScreen and jScan.
In addition to scleral images, the study explored using urine colour as a predictive marker. Despite the known correlation between jaundice and dark urine, predictions based solely on urine images were less reliable. Even when combined with scleral data, they did not enhance overall accuracy.
A key limitation was the relatively small dataset, which may not represent the full spectrum of clinical cases. Further validation with larger, more diverse data is needed to confirm the tool’s clinical utility.
Nonetheless, this study highlights the promise of simple, low-cost methods for early jaundice detection, especially when integrated into smartphone-based applications. By reducing the need for invasive blood tests, this AI-powered tool could significantly improve patient comfort and accessibility, especially in areas with limited medical resources.
Reference
Seok J et al. Non-invasive jaundice screening using AI: machine learning analysis of sclera and urine images. J Clin Med. 2025;14(9):3125. Published 2025 Apr 30. doi:10.3390/jcm14093125
