Hematological Classification of White Blood Cells by Exploiting Digital Microscopic Images

Abstract

A blood test is an essential examination process for evaluating body functions. Blood cell classification is an important laboratory process for detecting blood diseases. Microscopic evaluation by experts is a slow process, and the outcome depends on skill and experience. In addition, the process can be tedious and time-consuming. Therefore, an automated medical diagnostic system is essential in recognizing diseases in a short time and providing information about blood-related diseases such as leukemia. White blood cells are one of the most important types of blood cells associated with the immune system, as their forms are important and necessary for diagnosing blood diseases. In this study, a new Deep Learning (DL) network model called White Blood Cell Hematological Diseases Classification (WBC_HDC) is proposed. The classification was based on a convolution neural network (CNN) scheme for classifying hematology. A dataset containing 2800 images of white blood cells has been used, which were obtained with a CellaVision DM96 analyzer in the laboratory of the Barcelona Hospital Clinic (BHC). The data set is organized into the following four groups: Lymphocytes, Monocytes, Immature Granulocytes (IG), and Erythroblasts. The images' sizes are 360 x 363x 3 pixels in Joint Photographic Group (JPG ) format and have been annotated by clinical pathology experts. Images were taken from individuals without infection, hematology, or neoplasia and free of any drug treatment at the moment of blood collection. The WBC_HDC network has recorded an accuracy of 90.86% after going through numerous tests of network layer parameters for eight different stages.