Diffusion tensor fiber tracking based on unsupervised learning

Using Hebbian learning rule and its special case Self-Organizing Map (SOM) as unsupervised learning, a solution is proposed for defining the fiber paths which is a critical problem in diffusion tensor literature, and synthetic diffusion patterns are analyzed by artificial neural network (ANN) approach. Unsupervised learning in training neural networks is a method, where network classification rules are self developed and which does not require any knowledge about the desired output. Only input data is presented to the ANN in the learning process of the network, in other words the input space of the unsupervised learning ANN is the diffusion tensor eigenvector data of each imaging matrix. The network then adjusts the weightings to determine patterns having similar characteristics and classification is done in that way. The resulting classification represents the principal diffusion direction and the weighted diffusion distribution tracked by the fibers. Verification of the application on synthetic data enabled the implementation of the method on real brain images. The aim of the proposed method is to accomplish brain fiber tracking based on learning algorithms according to the modeling studies accepted in artificial neural network literature. Implementing SOM for fiber path discrimination purposes was successful and future work relies in 3D diffusion tensor tractography.