HOPFIELD NEURAL-NETWORK FOR THE MULTICHANNEL SEGMENTATION OF MAGNETIC-RESONANCE CEREBRAL IMAGES

In this paper, we present an approach for the segmentation of magnetic resonance images of the brain, based on Hopfield neural network. We f ormulate the segmentation problem as a minimization of an energy funct ion constructed with two terms, the cost-term, that is a sum of errors ' squares, and the second term is a temporary noise added to the cost- term as an excitation to the network to escape from certain local mini ma and be closer to the global minimum. Also, to ensure the convergenc e of the network and its utility in the clinic with useful results, th e minimization is achieved in a way that after a prespecified period o f time the energy function can reach a local minimum close to the glob al minimum and remain there ever after. We present here, segmentation results of two patients data diagnosed with a metastatic tumor and mul tiples sclerosis in the brain. (C) 1997 Pattern Recognition Society.