Leukocytes Segmentation and Classification in Digital Microscopic Images

Abstract

Image processing and machine learning have recently gained positive contributions to various medical procedures. One of the diagnostic processes' essential requirements in many diseases is laboratory tests, such as the Complete Blood Count (CBC) test. In CBC, various leukocytes, also known as White Blood Cells (WBC), are segmented, classified, and counted by a lab technician in microscopic slides. This process is very tiresome and requires a human technician with specialized skill sets. This research proposes a fully automatic algorithm for the segmentation and classification of white blood cells. The proposed method applies pre-processing techniques to digital microscopic images. White blood cells are then segmented based on color pallets. Hybrid features are extracted from the segmented images based on the fusion of local binary patterns and statistical features. Then various classifiers are used for the classification of WBC. Results suggest that the Support Vector Machine (SVM) and Artificial Neural Networks (ANN) outclass other classifiers. It was observed that the proposed methodology outperformed existing methods in terms of classification accuracy (97.5%).