NUCLEAR-MEDICINE IMAGE SEGMENTATION USING A CONNECTIVE NETWORK

A method for post-reconstruction nuclear medicine image segmentation b ased on an analogy to the Ising model of a two-dimensional square latt ice of N particles (pixels) is presented. A reconstructed 2-D slice im age is analyzed as a multi-pixel system where pixels correspond to a 2 -D lattice of points with non-zero interaction energy with their neare st neighbors. The model assumes that pixel intensities belonging to th e same homogeneous image region are relatively constant, where region intensity means (or labels) are determined by both statistical paramet er estimation and deterministic image analysis. The change in value of each pixel during the segmentation process depends on (1) the statist ical properties in the reconstructed image and (2) the states of its n earest neighbors. These changes are either in the direction of statist ically estimated intensity means or other previously analyzed regions of significance. The segmentation technique uses a new innovative rela xation labeling connective network. The global relaxation dynamics of the network are controlled by the interaction of local synergetic and logistic functions assigned to each pixel. This result may improve the localization of hot and cold regions of interest as compared to the o riginal image.