A comparative analysis of neural network methodologies for segmentation of magnetic resonance images

Presents a comparative study of the potential of artificial neural networks for the segmentation of multispectral MRI data. Multilayer perceptron (MLP) and radial basis function (RBF) neural networks have been applied to MR images as supervised segmentation methods. We have also applied analog adaptive resonance theory model (ART2) as an unsupervised segmentation method to MRI and have studied its function. We used identical data sets in this paper to compare the results. RBF proved the smallest execution time, but demonstrated more dependency on the training data than MLP. ART2 provided a good unsupervised technique for the MRI data