A decision support system for Acute Leukaemia classification based on digital microscopic images

Abstract

In the era of digital microscopic imaging, Image Processing, data analysis, classification, decision support systems have emerged as one of the most important tools for diagnostic research. Physicians can observe cellular internal structures abnormalities by visualizing and analyzing images. Leukemia is a malignant disease characterized by the uncontrolled accumulation of abnormal white blood cells. The recognition of acute leukemia blast cells in colored microscopic images is a challenging task. The first important step in the automatic recognition of this disease, image segmentation, is considered to be the most critical step. In this study, we present a decision support system that includes the panel selection, segmentation using K-means clustering to identify the leukemia cells and features extraction, and image refinement. After the decision support system successfully identifies the cells and its internal structure, the cells are classified according to their morphological features of this analysis the decision support system was tested using a public dataset designed to test segmentation techniques for identifying specific cells, and the results of this analysis were compared with those of other techniques, which were suggested by other researchers, applied to the same data. The algorithm was then applied to another dataset, extracted under the supervision by an expert pathologist, from a local hospital; the total dataset consisted of 757 images gathered from two datasets. The images of the datasets are labeled with three different labels, which represents three types of leukemia cells: blast, myelocyte, and segmented cells. The process of labeling of these images was revised by the expert pathologist. The algorithm testing using this dataset demonstrated an overall accuracy of 99.517%, the sensitivity of 99.348%, and specificity of 99.529%. Therefore, this algorithm yielded promising results and warrants further research.