Analysis and automated classification of images of blood cells to diagnose acute lymphoblastic leukemia

Abstract

Analysis of white blood cells from blood can help to detect Acute Lymphoblastic Leukemia, a potentially fatal blood cancer if left untreated. The morphological analysis of blood cells images is typically performed manually by an expert; however, this method has numerous drawbacks, including slow analysis, low precision, and the results depend on the operator’s skill. We have developed and present here an automated method for the identification and classification of white blood cells using microscopic images of peripheral blood smears. Once the image has been obtained, we propose describing it using brightness, contrast, and micro-contour orientation histograms. Each of these descriptions provides a coding of the image, which in turn provides n parameters. The extracted characteristics are presented to an encoder’s input. The encoder generates a high-dimensional binary output vector, which is presented to the input of the neural classifier. This paper presents the performance of one classifier, the Random Threshold Classifier. The classifier’s output is the recognized class, which is either a healthy cell or an Acute Lymphoblastic Leukemia-affected cell. As shown below, the proposed neural Random Threshold Classifier achieved a recognition rate of 98.3 % when the data has partitioned on 80 % training set and 20 % testing set for. Our system of image recognition is evaluated using the public dataset of peripheral blood samples from Acute Lymphoblastic Leukemia Image Database. It is important to mention that our system could be implemented as a computational tool for detection of other diseases, where blood cells undergo alterations, such as Covid-19