Leukocyte Classification using Adaptive Neuro-Fuzzy Inference System in Microscopic Blood Images

Microscopic pathology is still a meticulous and biased task for hematologist, which leads to the misclassification of cells and vagueness prediction of abnormal cells due to variability in the morphological structure of leukocytes. Therefore, to enhance the detection precision and diminishing the time factor, an automatic classification system for leukocytes has been proposed. In routine clinical practice, expert hematologists observed that the nucleus plays a crucial role in the identification of the blood disorders. Accordingly, in this work, the localization of leukocyte nucleus is performed by using Chan-Vase level-set method for the design of a classification framework that differentiates between four classes of the leukocytes, i.e., eosinophils, polymorphs, monocytes and lymphocytes based on the nucleus. A dataset consisting of 162 leukocyte microscopic images is used. The images in the dataset are classified on the basis of texture, shape and color features. The feature selection method based on the linguistic hedge is applied on evaluated feature space of 92. The selected features are fed to an adaptive neuro-fuzzy classifier for the classification. The proposed framework obtained an accuracy of 98.7% after applying the adaptive neuro-fuzzy classification on selected 46 informative features. The correlation of best features and data extorted from the different microscopic images may yield a dramatic increase in diagnostic consistency in clinical pathology. The results obtained by utilization of selected optimal features and adaptive neuro-fuzzy classification system indicate that it can be routinely used in clinical environment for differential diagnosis between different classes of leukocytes.