Automated detection and classification of leukaemia by image processing and deep learning techniques

Abstract

Leukaemia is a cancer that originates in the blood forming tissues of bone marrow newlineand results in large number of abnormal white blood cells (WBC) in the bone newlinemarrow and blood. These immature abnormal cells, known as blasts, crowd out newlinenormal blood cells and prevent their development. There are mainly four newlineclassifications for leukaemia: Acute Lymphoblastic Leukaemia (ALL), Acute newlineMyeloid Leukaemia (AML), Chronic Lymphocytic Leukaemia (CLL) and Chronic newlineMyeloid Leukaemia (CML). Leukaemia is primarily diagnosed based on the signs newlineand symptoms of the patient, Complete Blood Count (CBC) test and peripheral newlineblood smear examination by pathologists using a microscope. A bone marrow newlineexamination and advanced laboratory tests are also carried out to confirm and newlineclassify leukaemia. The conventional blood and bone marrow smear examination by newlinelight microscopy suffers from intra-observer and inter-observer variability. Image newlineprocessing-based techniques which can automatically analyse the images of blood newlineand bone marrow smears to identify abnormal cells can overcome these personal newlinebiases of the medical professionals. This study uses Deep Learning based newlineclassification techniques for computer aided detection and classification of newlineleukaemia. Detection of leukaemia in general is carried out in the study using pretrained Deep newlineNeural Networks (DNN) without using conventional image segmentation and newlinefeature extraction techniques which are computationally complex. The study uses a newlinepublicly available dataset ALL_IDB2 to carry out the classification of leukemic newlinecells against normal cells using three types of DNNs viz. Series, DAG and Residual. newline newline