Pixel-Based Artificial Neural Networks in Computer-Aided Diagnosis

Abstract

Artificial neural networks (ANNs) play an essential role in the medical imaging field, including medical image analysis and computer-aided diagnosis (CAD) (Doi 2005; Giger and Suzuki 2007), because objects such as lesions and organs in medical images may not be represented accurately by a simple equation. For example, a lung nodule is generally modeled as a solid sphere, but there are nodules of various shapes and nodules with internal inhomogeneities, such as spiculated nodules and ground-glass nodules. A polyp in the colon is modeled as a bulbous object, but there are also polyps which exhibit a flat shape (Lostumbo, Wanamaker et al. 2010). Thus, diagnostic tasks in medical images essentially require “learning from examples (or data).” One of the most popular uses of ANNs in medical image analysis is the classification of objects such as lesions into certain classes (e.g., abnormal or normal, lesions or non-lesions, and malignant or benign). The task of ANNs here is to determine “optimal” boundaries for separating classes in the multi-dimensional feature space which is formed by input features (e.g., contrast, area, and circularity) obtained from object candidates. Machine-learning algorithms for classification include linear discriminant analysis (Fukunaga 1990), quadratic discriminant analysis (Fukunaga 1990), multilayer perceptron (Rumelhart, Hinton et al. 1986), and support vector machines (Vapnik 1995). Such machine-learning algorithms were applied to lung nodule detection in chest radiography (Shiraishi, Li et al. 2006) and thoracic CT (Armato, Giger et al. 2001; Arimura, Katsuragawa et al. 2004), classification of lung nodules into benign or malignant in chest radiography (Aoyama, Li et al. 2002) and thoracic CT (Aoyama, Li et al. 2003), detection of microcalcifications in mammography (Wu, Doi et al. 1992), classification of masses into benign or malignant in mammography (Huo, Giger et al. 1998), polyp detection in CT colonography (Yoshida and Nappi 2001; Jerebko, Summers et al. 2003), determining subjective similarity measure of mammographic images (Muramatsu, Li et al. 2005; Muramatsu, Li et al. 2006; Muramatsu, Li et al. 2007), and detection of aneurysms in brain MRI (Arimura, Li et al. 2006). Recently, as available computational power increased dramatically, pixel/voxel-based ANNs (PANNs) emerged in medical image processing/analysis which use pixel/voxel values in images directly instead of features calculated from segmented regions as input