Analysis of Distance Transforms for Watershed Segmentation on Chronic Leukaemia Images

Abstract

Leukaemia is a blood cancer that contributes to the increase in the world mortality rates per year. Leukaemia can be divided into two major types which are acute and chronic leukaemia. This disease is caused by the excessive production of abnormal white blood cells (WBCs); hence these cells play a major role in the screening and diagnosis of leukaemia disease. Leukaemia screening requires the complete blood count process. However, due to the cells complex nature in chronic leukaemia which is overlapped, it would be difficult to obtain the accurate number of the WBCs for the screening process. Therefore, this paper proposes an automated WBCs counting with analysis of watershed segmentation for the screening of chronic leukaemia images. The segmentation approach consists of a few steps; (1) colour conversion, (2) image segmentation, (3) noise removal and (4) separation of overlapping WBCs. In this paper, three different distance transforms for watershed segmentation known as Euclidean, city block and chessboard have been analysed in order to find the best approach which is capable of separating the overlapping WBCs. The experimental results show that segmentation using watershed based on Euclidean has successfully segmented 50 blood images with average counting accuracy of 99.81%, as compared to the city block (91.09%) and chessboard (98.78%). Thus, the proposed procedures with watershed segmentation provide an efficient alternative in enhancing the accuracy of the WBCs count for leukaemia screening.