Newborn jaundice determination by reflectance spectroscopy using multiple polynomial regression, neural network, and support vector regression

Diffuse reflectance spectroscopy is a non-destructive method to obtain biochemical and physiological information by investigating the optical properties of skin. Transcutaneous bilirubin (TcB) measurement utilizes reflectance spectroscopy to determine the jaundice level in newborns. Although TcB measurement has some advantages over total serum bilirubin (TSB) measurement such as being non-invasive, noninfectious, painless, and instantaneous, the existing TcB devices cannot yet replace TSB devices due to the inaccuracy of measurements. In this paper, we propose the use of reflectance spectroscopy in conjunction with regression tools such as multiple polynomial regression (MPR), artificial neural network (ANN), and support vector regression (SVR) to predict the jaundice level. The proposed methods were tested on TcB measurement data obtained from 314 babies. TcB measurements were collected by two devices: a commercially available product, Draeger IM-103, and a prototype device on which we can implement the proposed algorithms. The results are encouraging towards increasing the clinical usage of transcutaneous bilirubinometers as all the three methods accurately predict the jaundice level with a correlation value between 0.932 and 0.943. The proposed use of ANN improves the non-invasive transcutaneous approach, with results converging to more accurate invasive serum bilirubin measurements by blood sampling. (C) 2019 Elsevier Ltd. All rights reserved.