Article Preview
TopIntroduction
Lung diseases are major threats because significant numbers of people suffer from these diseases such as Tuberculosis (WHO, 2014), pneumonia, lung cancer and pulmonary edema across the world. The advent of new powerful hardware and software techniques has triggered attempts to develop computer-aided diagnostic (CAD) systems for automatic chest x-ray screening (Karargyris, 2016) (Santosh, 2016) (Schaefer-Prokop, 2008). However, foreign element such as buttons on the gown that the patients were wearing or coins/buttons mistakenly swallowed by patients, within the chest x-ray images hinders the performance of the automatic screening process. Figure 1 shows a couple of such CXRs with circular foreign element (i.e., buttons) and Figure 2 shows a closer view of all the buttons in these CXRs. The presence of such element (especially the ones located within the lung region) hinders the CAD system performance, as they are not due to any lung abnormalities and therefore should not be considered. Therefore, in the screening process precise detection of foreign element is an important issue for screening of chest diseases in CAD system. In this paper, we focus on automatically detecting foreign circular element in CXR images.
In computer vision and image analysis, circle detection has a rich state-of-the-art (Chen & Wu, 2014) (Duda & Hart, 1972) (Rad et al., 2003) (Wu et al., 2013) (Zhou & He, 2015). Applications vary from document understanding: graphics recognition (Dosch et al., 2000) (Santosh, 2011) (Santosh et al., 2012) (Smith & Lamiroy, 2017) (Smith & Lamiroy, 2015) (Xu et al., 1990) to medical image analysis (Xue et al., 2015). Detecting circular foreign element in CXR images is a challenging and open research problem. However, in case of CXR images, we found (Xue et al., 2015) is only the work reported recently. In Xue et al., authors observed that CHT outperformed others but still, the technique is highly sensitive to intensity values. If there is not enough contrast between the element (buttons in CXRs) and the background image, then the performance of CHT degrades drastically. On the other hand, Viola-Jones algorithm has high false detection rate that can degrade the performance of the CAD system.
Figure 1. CXRs containing buttons, boxes marked by red indicate the location of buttons in the images
Figure 2. Closer view of all the buttons in above chest x-ray images