Self-adaptive RBF neural network-based segmentation of medical images of the brain

This paper proposes a method for segmentation of medical images of the brain by using a self-adaptive radial basis function neural network (RBF-NN), which imposes a confidence measure to select a subset of the RBFs in the hidden layer for producing outputs at the output layer, thereby making the network self-adaptive. This process reduces the computation time at the output layer of the RBF-NN by neglecting the ineffective RBFs and also it reduces the false recognition rate of the system. The centers of the different RBFs are identified by a modified version of the conventional k-means algorithm. A knowledge-based approach and point symmetry distance as similarity measure have been used in this algorithm to identify the centers of different RBFs of the network. The proposed method has been tested on both the simulated and real patient magnetic resonance (MR) and computed tomography (CT) images of the human brain and found to be better when compared with the approaches using the k-means, fuzzy c-means (FCM), and RBF-NN using conventional k-means algorithm to model the hidden layer neurons.