An Automated Multi-Level Convolutional Neural Network Approach for Classification of White Blood Cells

Blood cells known as white blood cells (or leukocytes) are generated in the marrow of the bone marrow and circulate through the circulatory system and lymph tissue. White blood cells (WBC) are the cells responsible for defending the body against outside invaders. They aid the body in its defence against infections and other illnesses. There are three types of white blood cells: granulocytes (which include basophils, eosinophils and neutrophils), monocytes, and lymphocytes (which include T cells and B cells). Usually, a total blood count test includes checking the blood's white blood cell count. Infection, inflammation, allergies, and leukaemia. White blood cells have a lifespan of 13 to 20 days before being eliminated in the lymphatic system. The first time immature WBCs leave the bone marrow and enter the peripheral circulation, they are known as bands or stabs. Phagocytosis is the method through which leukocytes combat infection. This work’s primary goal is to find the type of white blood cell from the provided input of white blood cells images and to categorize cell type according to its variety. Based on the progress made in developing the Deep Learning algorithm, this study has the potential to aid in the prediction of cell kinds and in assisting medical professionals in the diagnosis of a wide range of abnormalities and disorders. Inceptionresnetv2, VGG-16, CNN, are few deep learning model that can automatically assess white blood cell shots from different angles and circumstances of illumination in order to determine what kind of cell they are. The dataset that is used in the project contains four categories or classes with values ranging from 0 to 3, where 0,1,2 and 3 denotes particular variety of white blood cell. Among the models that have been utilised for the classification of images of blood cells, the accuracy obtained by proposed custom-built CNN framework is 98.70% that outperform the other framework.