Performance Comparison of Segmentation Techniques for Nucleus in Chronics Leukemia

Abstract

Morphological criteria have been used by haematologists to identify malignant cells in the blood smear sample under a light microscope. Experienced hematologist must perform this screening operation. However, manual screening using microscope is time-consuming and tedious. Thus, an automated or semi-automated image screening and diagnosis system are very helpful. An ideal automated screening system will acts as a human expert during the procedure. To formulate this idea, there are few steps involves in this process which is the acquisition of image, image segmentation, features extraction and recognition of image data for further analysis in computer-based. However, segmentation of a region of interest is the most crucial task to extract features for further learning and diagnose. This paper represents two segmentation techniques and their performance comparison based on clustering approach which are k-means and moving k-means clustering algorithms. The segmentation process is performed on ten chronics leukaemia images. The performance of segmentation based on the proposed techniques was evaluated. The proposed segmentation techniques offer high accuracies of segmentation which is more than 97% for both techniques.