Extensive Study of WBC Segmentation Using Traditional and Deep Learning Methods

Abstract

According to the WHO annual report, the death rate related to blood diseases is very high in the Asian continent. The existing traditional system is prolonged and tedious and also based on the expertise’s knowledge. Therefore, the development of an automated blood-related disorder diagnostic system is very essential to make the system error-free and more effective. As per the hematologist’s opinion, most of the disease can be identified by the White blood cells (WBC) related information. Thus, the main goal of this chapter is to segment WBCs from microscopic images using different traditional and deep learning (DL) algorithms. For this work, three traditional methods, i.e., global thresholding, k-means clustering and one DL-based model, i.e., U-Net is implemented. The extensive work of experimentations has been conducted on ALL-IDB dataset. The performance of the Global thresholding, the K-means clustering technique and the U-Net method is evaluated with the help of metrics like Jaccard Index, accuracy, sensitivity, and specificity. It has observed that the U-Net method is the best performing method for the WBC segmentation.