Automatic detection of tumors in the bile ducts of the liver is very difficult as often, in the defacto noninvasive diagnostic images using magnetic resonance cholangiopancreatography (MRCP), tumors are not clearly visible. Specialists use their experience in anatomy to diagnose a tumor by absence of expected structures in the images. Naturally, undertaking such diagnosis is very difficult for an automated system. This chapter proposes an algorithm that is based on a combination of the manual diagnosis principles along with nature-inspired image processing techniques and artificial neural networks (ANN) to assist in the preliminary diagnosis of tumors affecting the bile ducts in the liver. The results obtained show over 88% success rate of the system developed using an ANN with the multi-layer perceptron (MLP) architecture, in performing the difficult automated preliminary detection of the tumors, even in the robust clinical test images with other biliary diseases present.
TopIntroduction
There are a large number of algorithms and applications that have been and are actively being developed to assist in medical diagnosis. As medical problems are biological in nature, it is expected that nature-inspired systems would be appropriate solutions to such problems. Among the most popular of such nature-inspired tools is the artificial neural network (ANN), which has lent itself to applications in a variety of fields ranging from telecommunications to agricultural analysis. There is a large amount of literature on the use of ANN in medical applications. Some examples of medical systems developed employing neural networks include those for screening of heart attacks (Furlong et al., 1991) and coronary artery disease (Fujita et al., 1992), facial pain syndromes (Limonadi et al., 2006), diabetes mellitus (Venkatesan & Anitha, 2006), psychiatric diagnosis (NeuroXL, 2003), seizure diagnosis (Johnson et al., 1995), brain injuries (Raja et al., 1995), and many more.
There is an increasing number of cancer cases in most countries, with an increasing variety of cancers. Over the years, ANN has been actively employed in cancer diagnosis as well. ANN systems have been developed for cancer of the breast (Degenhard et al., 2002), skin (Ercal et al., 1994), prostate (Brooks, 1994), ovaries (Tan et al., 2005), bladder (Moallemi, 1991), liver (Meyer et al., 2003), brain (Cobzas et al., 2007), colon (Ahmed, 2005), lung (Marchevsky et al., 2004), eyes (Maeda et al., 1995), cervix (Mango & Valente, 1998) and even thyroid (Ippolito et al., 2004). ANN has also been used for cancer prognosis and patient management (Naguib & Sherbet, 2001).
Although there has been extensive development of ANN systems for medical application, there are still many more diagnostic systems for diseases and organs that would be able to gain from this nature-inspired technology. This chapter proposes a multi-stage nature-inspired detection scheme that mimics the radiologist’s diagnosis strategy, where most of the algorithms employed are themselves nature-inspired. The scheme is augmented with the nature-inspired neural networks to improve the system performance in tackling automatic preliminary detection of a difficult and much less researched set of tumors affecting the bile ducts, using the defacto diagnostic imaging technology for the liver and pancreato-biliary system.