Article Preview
Top1. Introduction
Brain tumor is one of the leading causes of mortality worldwide and an increasing threat in low and middle-income countries. In clinical, brain tumor occupies the 13th place in frequency of all cancers. Moreover, Brain tumor is the second most common cancer in children and the most common cause of cancer death in children. Clinical studies show that the chance of survival could be increased if the brain tumor is detected correctly at its early stage. Thus, Patients with the brain tumor may benefit from early diagnoses in order to receive primary care or undergo surgical procedures to address their progressive symptoms. A robust method which could offer quantitative measurements for brain tumors and disease patterns in digital pathology images to achieve accurate segmentation results, however, remains challenging (Tohka, 2014; Wang et al., 2011; Hu et al., 2010; Weiner et al., 2013).
In medical community, there are the limitations for radiologists to detect and diagnose diseases in digital pathology images both by their non-systematic search patterns and by the presence of image noise during image pattern interpretation. On the other hand, similar characteristics of some abnormal and normal structure may cause interpretational errors in images. In addition, the huge amount of image data produced by imaging devices, for example, microscopic imaging and magnetic resonance imaging, makes the detection of potential diseases a burdensome task causing oversight errors in digital pathology images. Developments in computer vision and image processing algorithms in medical image interpretation have shown that the computer aided diagnosis system can pursue the major objective of carrying out mass screening campaigns, acting as a fully automated system (Tohka, 2014; Wang et al., 2011; Ahmed et al., 2015). This study aims to assist radiologists in identifying difficult cases in digital pathology images, when such conflicts of cancer nuclei involve ‘omics’ indicative of brain tumors pathologically. As the segmentation of cancer nuclei is crucial early steps on supporting of brain tumor detection, it has become one of the important areas of research in assess patients with brain tumor in medical community (Ahmed et al., 2015; Duchesne et al., 2008; Paul et al., 2015).
Image segmentation refers to the partitioning of an image into disjoint regions with respect to a chosen property such as texture and color. In the image segmentation stage, many popular methods relied on the pre-processing steps for the task of disease pattern detection in clinical decision support systems (El-Zehiry et al., 2007; Greenspanm et al., 2011; West et al., 2012). Alternatively, another popular category of image segmentation methods is based on the thresholding method (Otsu, 1979; Tohka, 2014), utilizing geometric information such as deformable models using a minimization of an energy functional (Jack et al., 2010; Yushkevich et al., 2006), the graph cut optimization (PePe et al., 2013), and the watershed transform algorithms (Duyn, 2012; Weiner et al., 2013). However, in the segmentation, the watershed transform algorithms were performed on the intensity inverted image by selecting the basin to represent the brain. In addition, the deformable models required appropriate parameter settings and considerably complex pre-processing steps in image segmentation for clinical decision support systems. A review of some methods can be found in the work (Tohka, 2014; Weiner, et al., 2013; Liew, et. al., 2006).