Fat Droplets Identification in Liver Biopsies using Supervised Learning Techniques

Abstract

Non-Alcoholic Fatty Liver Disease (NAFLD) is a frequent syndrome that exclusively refers to fat accumulation in liver and steatohepatitis1. It is considered as a massive disease ranging from 20% to 40% in adult populations of the Western World. Its prevalence is related to insulin resistance, which places individuals at high rates of mortality. An increased fat accumulation rate, can significantly increase the development of liver steatosis, which in later stages may progress into fibrosis and cirrhosis. In recent years, research groups focus on the automated fat detection based on histology and digital image processing. The current project, extends our previous work for the detection and quantification of fatty liver, by characterizing histological findings. It is an extensive study of supervised learning of fat droplet features, in order to exclude other findings from fat ratio computation. The method is evaluated on a set of 13 liver biopsy images, performing 92% accuracy.