| Felix Outlaw1, Miranda Nixon1, Terence S Leung1, Nicola Robertson2, Nana Okai Brako3, Judith Meek4, Christabel Enweronu-Laryea5 |
| Institution(s) |
| 1Dept of Medical Physics and Biomedical Engineering, UCL; 2UCL Institute for Women’s Health and UCLH; 3Greater Accra Regional Hospital; 4The Neonatal Care Unit, Elizabeth Garrett Anderson Wing, UCLH; 5Dept of Child Health, University of Ghana School of Medicine and Dentistry and Korle-Bu Teaching Hospital |
| Introduction (include hypothesis) |
| Timely identification of jaundice remains problematic. Specialised screening devices such as the transcutaneous bilirubinometer exist but are rarely available in the home setting, particularly in low-income parts of the world. In Sub-Saharan Africa, long-term impairment following kernicterus is eight times more common than in high-income countries [1]. Smartphones may serve as objective and ubiquitous colorimeters. We seek to show that a sclera imaging app could screen for jaundice with comparable accuracy to a transcutaneous bilirubinometer. |
| Methods |
| The app was used to image the sclera of 101 newborns in the Greater Accra Regional Hospital before a blood test for determination of total serum bilirubin (TSB) concentration. Pixel values were extracted from manually drawn regions of interest in the sclera. A regression model was used to relate sclera chromaticity to TSB. We call the output of this model the Scleral-Conjunctival Bilirubin (SCB), in analogy with Transcutaneous Bilirubin (TcB). A TcB measurement was also made using the Draeger JM-105. This study is funded by Saving Lives at Birth with ethical approval from Ghana Health Services, UCL, and University of Ghana Ethics Committees. |
| Results |
| A correlation of r=0.79 (p<0.01) was achieved between SCB and TSB. In screening for newborns with TSB greater than 250μmol/L, the NICE-recommended threshold for blood test referral in term infants [2], a SCB decision threshold of 215μmol/L maximised Youden’s Index. The sensitivity (95% confidence interval) was 96% (89% – 99%), with specificity 72% (62% – 81%). The area under the Receiver Operating Characteristic curve (AUROC) was 0.91. With the same subjects and decision criteria, the JM-105 gave correlation 0.76 (p<0.01), sensitivity 89% (81% – 94%), specificity 77% (67% – 85%), and AUROC 0.84. |
| Conclusions |
| We have shown that a colour-based jaundice metric from a sclera image has a screening performance at least as good as the latest generation of TcBs, and that this imaging is feasible with a smartphone in a clinical setting in Ghana. We will integrate analysis into the app for real-time results and improve robustness and usability. |
| References |
| [1] Bhutani, Vinod K., et al. Pediatric research 74.S1 (2013): 86. [2] National Institute for Clinical Excellence. “Jaundice in newborn babies under 28 days.” (2016). |