[Evaluation of an automatic image-based screening technique for neonatal hyperbilirubinemia].

OBJECTIVE: To verify the accuracy of a smart phone software application, which is an automated image-based bilirubin (AIB) testing technique for testing peripheral blood total bilirubin (TB) in term and late preterm neonates.

METHOD: During July 2015 to January 2016, jaundiced neonates were enrolled from department of neonatology in Tongji Hospital, Children's Hospital of Zhejiang University School of Medicine and Shenzhen Bao'an Maternal and Child Health Hospital. The enrolling criteria included gestational age (GA)≥34 weeks and peripheral blood total bilirubin test is required. Near-simultaneous TB, transcutaneous bilirubin (TcB), and AIB were determined. AIB consists of a cloud based off-line learning module and an on-line prediction module. Skin image, by smart phone is uploaded to a specific cloud server that includes learning based software to provide an individualized and immediate predicted bilirubin index. The t-test was used to determine if AIB and peripheral blood TB are significantly different. Agreement of AIB and TcB were compared with peripheral blood TB by Bland-Altman analysis. Linear regression was adopted to model the relationship of AIB and peripheral blood TB. The ROC curve of AIB was also plotted.

RESULT: Two hundred and fifteen neonates were enrolled, gestational age 37 weeks+ 5 d, postnatal age (11±4) d, weight (2.9±0.7) kg, 116 male and 99 female. One hundred and forty-eight term neonates (GA≥37 weeks) and 67 preterm neonates (GA 34-<37 weeks) were enrolled in this study. There was no significant difference between AIB (197±51)μmol/L and peripheral blood TB(191±65)μmol/L (t=1.611, P=0.109). There was strong relevance between peripheral blood TB and AIB as shown by Bland-Altman analysis (96% (207/215) samples lay within the 95% limits of agreement). The regression analysis showed that the r(2) was 0.593 in the whole population, while the r(2) was 0.628 in the term neonates. The ROC of AIB yielded a 0.743 AUC, and with 82% sensitivity and 60% specificity based on Youden index criterion.

CONCLUSION: Based on AIB's agreement with peripheral blood TB, the sensitivity and specificity, AIB can be used as a new technique to provide results for objective follow-up for progression and regression of jaundice.