Öz düzenleyici haritalar ile difüzyon tensör yolak takibi

The diffusion tensor imaging (DTI) is unique in its ability to estimate the white matter (WM) fiber tracts in vivo noninvasively. The post-processing of DT images needs proper image analysis and visualization tools. However, accurate WM anatomical maps should be provided to clarify the multiple orientational fiber paths within uncertainty regions. These regions with intersecting trajectories generate a critical tractography issue in DTI literature. WM fiber tractography needs a standardization, a generally accepted fiber tract atlas which is the main concern of the various research groups in the field. In this thesis, the special class of artificial neural networks (ANN) namely Kohonen's self organizing feature maps (SOFMs) is proposed for the analysis of DT images. This SOM based tractography approach called SOFMAT (Self-Organizing Feature Mapping Tractography) relies on unsupervised learning method for the mapping of high dimensional data into a 1D, 2D, or higher dimensional data space depending on the topological ordering constraint. The unsupervised approach enables SOFMAT to order the principal diffusivity of the fibers in the DTI into neural pathways. A major advantage of the topological maps produced by SOFMAT is that it retains the underlying structure of the input space, while the dimensionality of the input space is reduced. As a result, an artificial neuronal map is obtained with weights encoding the stationary probability density function of the input pattern vectors. Building fiber tracking maps based on the diffusion tensor information which learn through self organization in a neurobiologically aspect is the aim of the study. SOFMAT has been tested to reveal uncertainties in fiber tracking. A well known artificial dataset called PISTE was used to access the capabilities of SOFMAT. After identifying an affective configuration, SOFMAT was employed for human tractography.