Multi-class ECOCAMD Classifier in Classification of the types of White Blood Cells

Abstract

The primary hematological malignancy leukemia spoils the immune system of the body by changing the basic characteristics of White Blood Cells (WBC); consequently, they grow up very rapidly in the count.  Timely detection of leukemia and its types can be helpful to better diagnose the disease and treat accordingly. For the identification of types of leukemia, it is must to discriminate the types of WBC. This work is an effort to propose a method for automating the segmentation task of the nucleus region of WBC using YCbCr color space and local thresholding with 100% of segmentation accuracy. After this, the necessary set of statistical and texture features are extracted and given to the proposed machine learning multi-class Error Detection and Correction Code with Accurate Mean and Distance (ECOCAMD) classifier to classify the types of WBC. The results of this proposal demonstrate 98.81% of overall accuracy rate with 242 test images of blood smears obtained from the LISC dataset.