An Image Processing-based Tool for Noninvasive Measurement of Bilirubin Level

Abstract

Jaundice is a yellowish pigmentation of the skin, the conjunctiva, and other mucous membranes caused by increased levels of bilirubin in blood beyond normal levels (hyperbilirubinemia). Jaundice occurs in neonates in 2 forms: Pathological and Physiological Jaundice. In Pathological Jaundice, high bilirubin levels can result in severe neurotoxicity, brain damage, hearing loss, muscular disorders and physical abnormalities. So measuring the exact level of bilirubin in blood in neonates is very crucial. In the past, serum bilirubin has been the preferred method of detecting hyperbilirubinemia in newborns. The ordering of serum bilirubin in neonates is based on visual evaluation by either physicians or nursing staff. Skin puncture collection of blood exposes the neonate to trauma and risk of infection. A noninvasive device for predicting serum bilirubin levels in newborns diminishes the need to do skin punctures. There are numerous studies demonstrating the transcutaneous bilirubin measurements reasonably close to serum bilirubin performed by clinical laboratories. These devices use spectrophotometry (light absorption measurements at standardized wavelengths of light) to calculate measurements. All existing methods involve transcutaneous detection of bilirubin, which is not error free, as there are various factors affecting its accuracy, like presence of hemoglobin and melanin, and maturity of skin tissue and fibres. So this gave an urge for us to develop a new technique of quantitating bilirubin level noninvasively. This study was done by analyzing conjunctival images that would imitate the blood level of bilirubin by calculating the intensity of yellow color. Machine learning techniques were applied in order to develop a system capable of predicting the bilirubin level.