Abstract
Acute myeloid leukemia (AML) is a cancer of the myeloid line of cells caused due to the rapid increase of abnormal cells that later interfere with healthy cells. One of the main reasons for the increase in mortality is the cost of the devices used for the determination and late diagnosis. The most effective treatment option can be provided by accurate medical diagnosis. Automated segmentation of blood smear images plays a crucial role in the identification of the AML. This article proposes a new computer-aided diagnosis model to segment the blood smear images and identifies the stage of AML. The methodology presented in this work consists of various stages: Image acquisition, image segmentation, feature extraction/selection, and classification. The model is trained using 800 blood smear images collected from Kasturba Medical College Manipal, and 200 images collected from the dataset of microscopic peripheral blood cell images for the development of automatic recognition systems. The model is tested on 500 images. A novel algorithm is designed to accurately segment the blood smear images to identify AML and its stages. The segmentation algorithm addresses critical issues in blast cell detection, including identifying the blast cell and extracting the cytoplasm of the cell without involving manual intervention. It can identify the multiple lobes and the nucleated red blood cells (NRBCs), separate the overlapped erythrocytes from white blood cells (WBCs), and discover the presence of Auer rods and granules. The feature selection is performed using the InfoGainAttributeEval and the ranker search method. This article compares the performance of the various machine learning algorithms exploited for the classification of different types of cells and hence determines AML. The model successfully differentiated between NRBC and WBC with an accuracy of 99.81%. The model obtained a classification accuracy of 99.48 %. It achieved a prediction accuracy of 99.2% while predicting the unknown stage of AML. The algorithm's efficiency proves that it can be used by pathologists to form a prognosis.