Two-layer competitive based Hopfield neural network for medical image edgedetection

In medical applications, the detection and outlining of boundaries of organ s and tumors in computed tomography (CT) and magnetic resonance imaging (MR I) images are prerequisite. A two-layer Hopfield neural network called the competitive Hopfield edge-finding neural network (CHEFNN) is presented for finding the edges of CT and MRI images. Different from conventional 2-D Hop field neural networks, the CHEFNN extends the one-layer 2-D Hopfield networ k at the original image plane a two-layer 3-D Hopfield network with edge de tection to be implemented on its third dimension. With the extended 3-D arc hitecture, the network is capable of incorporating a pixel's contextual inf ormation into a pixel-labeling procedure. As a result, the effect of tiny d etails or noises will be effectively removed by the CHEFNN and the drawback of disconnected fractions can be overcome. Furthermore, by making use of t he competitive learning rule to update the neuron states, the problem of sa tisfying strong constraints can be alleviated and results in a fast converg ence. Our experimental results show that the CHEFNN can obtain more appropr iate, more continued edge points than the Laplacian-based, Marr-Hildreth, C anny, and wavelet-based methods. (C) 2000 Society of Photo-Optical instrume ntation Engineers. [S0091-3286(00)01903-6].