MR brain image classification by multimodal perceptron tree neural network

We propose a multimodal perceptron tree (MMPT) neural network to segment magnetic resonance (MR) images. The architecture consists of simple networks-neurons, hierarchically connected in a tree structure. The latter is built up during training by the adopted depth-first searching technique augmented with choosing the best hyperplane split of the feature subspace at each tree node. This neural network effectively partitions the feature space into subregions and each terminal subregion is assigned to a class label depending on the data routed to it. As the tree grows, the number of training data for each node decreases, which results in less weight update epochs and decreases the time consumption. The MMPT performance is compared to that of a multilayered perceptron (MLP). The networks are applied to brain MR image segmentation into gray matter/white matter regions.