Smartphone Imaging and Artificial Intelligence For Neonatal Jaundice Screening: From Colorimetry To Clinical Validation

Abstract

Neonatal jaundice is common during the newborn period, but clinically significant hyperbilirubinemia requires timely detection and appropriate follow-up. Total serum bilirubin (TSB) remains an important reference measurement, while transcutaneous bilirubinometry (TcB) provides a non-invasive screening approach. Smartphone imaging combined with artificial intelligence (AI) has emerged as a potential complementary method for bilirubin assessment.

This article synthesizes published evidence on smartphone-based neonatal jaundice screening, covering colourimetric approaches, machine learning, deep learning, calibration, population diversity and clinical implementation. Findings from studies of BiliCam, Biliscan, Picterus, BiliSG and other smartphone-based systems indicate that image-derived measurements can be associated with TSB. However, reported performance varies with the study population, imaging device, acquisition conditions, bilirubin range and validation method. Important unresolved issues include underestimation at high bilirubin concentrations, performance in preterm infants and populations with darker skin pigmentation, cross-device generalizability, image-quality failures and limited prospective external validation.

The evidence also highlights the need to distinguish bilirubin estimation from threshold classification, referral decisions and demonstrated clinical benefit. Future investigations should prioritize multicentre datasets, standardized acquisition and calibration, subgroup-specific performance assessment, uncertainty-aware prediction and prospective evaluation of screening pathways. Smartphone-based systems should currently be considered potential screening and decision-support tools rather than replacements for established bilirubin measurement.