Non-invasive early diagnosis of jaundice with computer vision

Abstract

Jaundice is a condition characterized by the yellowing of skin and sclera of the eyes. Jaundice occurs when the liver is unable to eliminate the bilirubin, a waste material formed as a result of the breakdown of red blood cells. The excessive accumulation of bilirubin in the blood can result in permanent brain damage. Therefore, Jaundice has to be identified in the early stage. The diagnosis method available is a bilirubin blood test, a painful procedure where blood sample is collected from patient. To mitigate the pain, developing an alternative non-invasive approach can aid in the early diagnosis of jaundice. A lot of research has been carried out to develop a non-invasive procedure. Most of the works focused on identifying jaundice by analysing the yellowness of skin. However, yellowing of the skin is a less noticeable symptom if the patient has darker skin, but the yellowing of the sclera can be more easily identifiable. This work focuses on identifying jaundice from the sclera. There is no standard publicly available dataset for jaundice diagnosis and all the previous works are carried out by collecting data from hospitals. One of the major limitations of medical data is its limited availability. In Artificial Intelligence (AI) research insufficient data results in incorrect predictions. However, previous works in this area have not worked on increasing the volume of the dataset. Through this work, the potential of the Generative Adversarial Network (GAN) is leveraged to overcome the issue of limited medical data availability. In this paper, a hybrid approach based on computer vision and classical machine learning is developed that can accurately determine the intensity of jaundice from the yellowness of the sclera. The work addresses the challenges of limited medical dataset availability while considering the privacy of the concerned individual.

Topics & Concepts