The grey-white decision network is a constraint satisfaction neural network designed to perform low-level segmentation of magnetic resonance brain images. The GWDN assigns a grey matter, white matter, or ``other'' matter label to each voxel using constraints imposed by edges, neighbors, signal intensity, and the need to make a distinct choice between expected tissue classes.
As part of an ongoing project for automatic segmentation of magnetic resonance (MR) brain images, we introduced the grey-white decision network (GWDN) in Worth91 as a three layer dynamical system that relaxes into a solution of the segmentation problem. A more complete description is described in Worth92.
The original GWDN was re-cast as a recurrent competitive field (RCF) instead of the original recurrent cooperative/competitive field (RCCF) so that it is stabile according to the Cohen-Grossberg theorem (Cohen and Grossberg, Cohen83). The new network is described in Worth94 and not only has improved mathematical properties, but it is also faster. This paper contains the most complete description of the network. Prof. Siegfried Stiehl and his students, Thorsten Lange, Sabine Reimers, and Andreas Heinrich at the University of Hamburg have examined and extended the GWDN (Lange, Reimers, and Heinrich, Lange93, Lange94).
The most recent work done at the CMA on the GWDN has not been published. A postscript version of the latest developments can be had by clicking here. Changes were made to allow more robust parameter selection and test images were devised to set parameters.
Michael Farmer uses a Recurrent Competitive Field Neural Network in his Masters Thesis as described in Farmer96.
You want source code? For what it's worth, you can have an old version that I once cleaned up so I could make it available. Also, Michael Farmer has made his code available.
If you have any comments or questions, please ask me.
Worth, A.J., Lehar, S., and Kennedy, D.N. (1991) A recurrent cooperative/competitive field for segmentation of magnetic resonance brain imagery, Proceedings of the International Joint Conference on Neural Networks, November (Singapore), 2, 1403-1408.
Worth, A.J., Lehar, S., and Kennedy, D.N. (1992) A recurrent cooperative/competitive field for segmentation of magnetic resonance brain images, IEEE Transaction on Knowledge and data Engineering, Vol. 4, No. 2, 156-161.
Worth, A.J. and Kennedy, D.N. (1994) Segmentation of magnetic resonance brain images using analog constraint satisfaction neural networks, Image and Vision Computing, Vol. 12, No. 6, July/August, pp. 345-354.
Lange, T., Reimers, S., and Heinrich, A. (1993) Untersuchung eines kuenstlichen neuronalen netzes zur segmentierung von MR-Bildern. Studienarbeit, Fachbereich Informatik, Arbeitsbereich Kognitive Systeme, Universitaet Hamburg, Hamburg, Germany.
Lange, T., Reimers, S., and Heinrich, A. (1994) Untersuchung eines neuronalen Netzes zur Segmentierung von Kernspintomogrammen und Entwicklung einer Evaluierungsstrategie. Diplomarbeit, Fachbereich Informatik, Arbeitsbereich Kognitive Systeme, Universitaet Hamburg, Hamburg, Germany.
Cohen, M.A. and Grossberg, S. (1983) Absolute stability of global pattern formation and parallel memory storage by competitive neural networks, IEEE Trans on Sys. Man and Cyber. 13(5), 815-826.
Farmer, M.W., (1996) "Automatic Segmentation of Medical Images using Fuzzy C-Means Clustering, Oriented Edges, and a Recurrent Competitive Field Neural Network," Masters Thesis, University of Tennessee at Chattanooga, June.