Article Preview
Top1. Introduction
Cervical cancer is one of the most common threats for women worldwide, as it has been the second highest cause of cancer-caused deaths among women (Smith et al, 2014). Pap smear examination is one of the most common methods used to detect cervical cancer. The specific examination is a preventative measure to detect the presence of pre-cancerous and cancerous situations in cervical cell samples. The diagnostic value of the Pap smear is verified by the fact that cervical cancer incidence and mortality rates have declined since the introduction of the Pap smear test in the mid-20th century (Papanicolaou, 1942), and the rates continue to decline till now (Smith et al, 2014).
Pap smear images present particular characteristics, and their automated interpretation is still a challenging issue for the researchers. The diversity of the cell structures, the intense fluctuation of the background and the variances in illumination and dye concentration of the cells due to the staining procedure are some of the most representative limitations that any automated method should effectively deal with.
The presence of inflammatory cells and blood in the background is a common feature of conventional Pap smear images. In Figure 1 some examples of the inflammatory and overlapping cells are depicted. From these examples it is clear that an image analysis technique should be firstly performed in order to achieve a reliable interpretation about the contents of each image. This technique could include a number of steps, such as a) inflammatory cell extraction, b) segmentation of normal cells and abnormal cells, c) identification of menopause cells due to the increasing level of hormone, d) background cleaning due to the existence of blood cells. This work is focused on the extraction of inflammatory cells.
Figure 1. Conventional Pap smear cell images with inflammatory cells
Although there are plenty of methods proposed in the literature for the analysis of Pap smear images, the problem of inflammatory cell identification has not been addressed yet. More specifically, the determination of the nucleus and cytoplasm borders in cervical images that contain only one cell or isolated cells has been considered by several researchers (Bamford et al, 1996, Bamford et al, 1998, Lassouaoui et al, 2003, Yang-Mao et al, 2008, Lin et al, 2009). The methods proposed in (Plissiti et al, 2011(a), Plissiti et al, 2011(b)) deal with the detection of nuclei locations and the nuclei boundary delineation respectively, in conventional Pap-stained cervical cell images which may contain both isolated cells and cell clusters.
Furthermore, many methods do not directly use the color information of cervical images. In (Garrido et al., 2000) a method using grayscale images which affects the excess of edge points or overlapped objects in complex images was proposed. In addition, genetic algorithms (Lassouaoui et al, 2003), pixel classification (Baak et al, 2004), region growing (Mat Isa, 2005) and deformable models (Plissiti et al., 2010) contour detectors (Tsai et al, 2008; Malm & Brun, 2009) were also proposed for the segmentation of cervical images using grayscale images. It must be noted that neither of the above mentioned studies deal with the existence of inflammatory cells in Pap smear images.