CT image labeling using Hopfield neural network
Domagoj Kovačević, Sven Lončarić · 2002
A method for the computed tomography (CT) image labeling is presented. CT images used in this work are obtained from patients having the spontaneous intra-cerebral haemorrhage (ICH). The images are segmented into three tissue classes (skull, brain, and ICH) and the background. The method consists of two steps. In the the first step, the image is divided into a number of regions using the K-means clustering algorithm. Regions used are dark, medium dark and bright region. In the second step, the regions are labeled using the modified Hopfield (1985) neural network. The stable state of the network represents a possible solution to the labeling problem. Simulated annealing is used as algorithm for network simulation.