Detection and Classification of White Blood Cells through Deep Learning Techniques

Samir Abou El-Seoud,Gerard Mckee,Muaad Hammuda Siala

doi:10.3991/ijoe.v16i15.15481

Abstract

Leukemia is one of the deadliest diseases in human life, it is a type of cancer that hits blood cells. The task of diagnosing Leukemia is time consuming and tedious for doctors; it is also challenging to determine the level and type of Leukemia. The diagnoses of Leukemia are achieved through identifying the changes on the White blood Cells (WBC). WBCs are divided into five types: Neutrophils, Eosinophils, Basophils, Monocytes, and Lymphocytes. In this paper, the authors propose a Convolutional Neural Network to detect and classify normal white blood cells. The program will learn about the shape and type of normal WBC by performing the following two tasks. The first task is identifying high level features of a normal white blood cell. The second task is classifying the normal white blood cell according to its type. Using a Convolutional Neural Network CNN, the system will be able to detect normal WBCs by comparing them with the high-level features of normal WBC. This process of identifying and classifying WBC can be vital for doctors and medical staff to make a decision. The proposed network achieves an accuracy up to 96.78% with a dataset including 10,000 blood cell images.

Highlights

Leukemia is one of the deadliest diseases threatening humanity [1]
The accuracy rate of the recognition of types of white blood cells by our proposed Convolutional Neural Network model achieved 96.78%
Due to its efficiency in diagnosing Leukemia at early stages, the convolutional neural network would be of a great value to the medical diagnostic system used to detect Leukemia in the blood cells

Summary

Introduction

Leukemia is one of the deadliest diseases threatening humanity [1]. It is a form of cancer that hits blood and bone marrow. The most common algorithms used to detect and classify microscopic images are the following methods: pre-processing; clustering; morphological filtering; segmentation; feature extraction, classification, and evaluation [7] These common methods have many disadvantages including long development time, the selection of the features in order to obtain best accuracy; it is difficult sometimes to make an accurate decision on whether the cell is abnormal or not. This paper will propose an application using a convolutional neural network, that detects white blood cells from microscope images and classifies these blood cells into one of the four classes: Class A: Monocytes, Class B: Lymphocytes, Class C: Neutrophils, Class D: Eosinophils. This paper will propose an application using a convolutional neural network, that detects white blood cells from microscope images and classifies these blood cells into one of the four classes: Class A: Monocytes, Class B: Lymphocytes, Class C: Neutrophils, Class D: Eosinophils. l

Related Work

Proposed Solution

Part 1: Features Extraction Part 2

Results

Conclusion

Authors