Detection of Acute Leukemia Using Segmentation and Classification Algorithm

Abstract

Blood is a body fluid, which is composed of blood cells (white and red blood cells), along with platelets suspended in blood plasma. In the medical field, segmentation and extraction of blood smear cells is very important. Leukemia is a blood cancer that mainly categorized into two types acute and chronic. In this research, we focused on to identifying Acute Lymphocytic Leukemia (ALL), and Acute Myelocytic Leukemia (AML) as leukemia type blood cancer. ALL appears in the blood cells due to abnormal growth of immature WBCs in the bone marrow. AML starts growing from the soft inner most part of bone. Both mainly affect children and people above 50 years. The mortality rate has risen sharply due to the late diagnosis and cost of the devices used for the diagnosis purpose. The flow cytometry method, which performs automatic counting of blood cells, fails to detect abnormal cells. Recalculations using a hem cytometer are prone to errors and are inaccurate. In this paper, a segmentation and classification based model for detecting ALL, and AML type leukemia is presented. The morphological based pre-processing technique is used to visualize the exact blast cell in the image by minimizing the normal cell visibility. After that, to find out the exact region of blast cells K-means is used for segmentation. To overcome the problem of foreground (blast cell) and background (normal cell) pixel mix up, Firefly Algorithm (FFA) as Swarm Intelligence (SI) algorithm is used. As the proposed algorithm, PSO is hybridized with FFA as a fitness function to FFA. To extract the features from the segmented blast cells, a Histogram of Oriented Gradients (HOG) descriptor is used. The detection of blood smear cells is performed using K-means, K-means with FFA, and K-means with Particle Swarm Optimization (PSO) approach. This is performed to know the best combination of SI with the K-means technique. AT last, the concept of Machine Learning (ML) using Artificial Neural Network (ANN) is used to train the proposed design automated leukemia segmentation and classification model that helps to classify the types of lymphoblastic leukemia in terms of ALL and AML with the performance parameters. The result shows that ANN with FFA offered an accuracy of 96.6%, and outperformed compared to existing work.