Automated Leukemia Screening and Sub-types Classification Using Deep Learning

Abstract

Leukemia is a kind of blood cancer that damages the cells in the blood and bone marrow of the human body. It produces cancerous blood cells that disturb the human’s immune system and significantly affect bone marrow’s production ability to effectively create different types of blood cells like red blood cells (RBCs) and white blood cells (WBC), and platelets. Leukemia can be diagnosed manually by taking a complete blood count test of the patient’s blood, from which medical professionals can investigate the signs of leukemia cells. Furthermore, two other methods, microscopic inspection of blood smears and bone marrow aspiration, are also utilized while examining the patient for leukemia. However, all these methods are labor-intensive, slow, inaccurate, and require a lot of human experience and dedication. Different authors have proposed automated detection systems for leukemia diagnosis to overcome these limitations. They have deployed digital image processing and machine learning algorithms to classify the cells into normal and blast cells. However, these systems are more efficient, reliable, and fast than previous manual diagnosing methods. However, more work is required to classify leukemia-affected cells due to the complex characteristics of blood images and leukemia cells having much intra-class variability and inter-class similarity. In this paper, we have proposed a robust automated system to diagnose leukemia and its sub-types. We have classified ALL into its sub-types based on FAB classification, i.e., L1, L2, and L3 types with better performance. We have achieved 96.06% accuracy for subtypes classification, which is better when compared with the state-of-the-art methodologies.