Computer-Aided Acute Leukemia Blast Cells Segmentation in Peripheral Blood Images

Computer-aided diagnosis system of leukemic cells is vital tool, which can assist domain experts in the diagnosis and evaluation procedure. Accurate blast cells segmentation is the initial stage in building a successful computer-aided diagnosis system. Blast cells segmentation is still an open research topic due to several problems such as variation of blats cells in terms of color, shape and texture, touching and overlapping of cells, inconsistent image quality, etc. Although numerous blast cells segmentation methods have been developed, only few studies attempted to address these problems simultaneously. This paper presents a new image segmentation method to extract acute leukemia blast cells in peripheral blood. The first aim is to segment the leukemic cells by mean of color transformation and mathematical morphology. The method also introduces an approach to split overlapping cells using the markercontrolled watershed algorithm based on a new marker selection scheme. Furthermore, the paper presents a powerful approach to separate the nucleus region and the cytoplasm region based on the seeded region growing algorithm powered by histogram equalization and arithmetic addition to handle the issue of non-homogenous nuclear chromatin pattern. The robustness of the proposed method is tested on two datasets comprise of 1024 peripheral blood images acquired from two different medical centers. The quantitative evaluation reveals that the proposed method obtain a better segmentation performance compared with its counterparts and achieves remarkable segmentation results of approximately 96% in blast cell extraction and 94% in nucleus/cytoplasm separation.