Segmentation of magnetic resonance brain images using analogue constraint satisfaction neural networks

摘要

The Grey-White Decision Network (GWDN) is presented as an analogue constraint satisfaction neural network that segments magnetic resonance brain images. Constraints on signal intensity, neighbourhood interactions and edge influences are combined to assign labels of grey matter, white matter or ‘other’ to each pixel. An improved version of this novel segmentation network that is provably stable is described. Results of the network are presented, along with a comparison of these results to a collection of human segmentations. The network is discussed in relation to other methods for segmentation, and the network's extendibility is described.