Automatic Segmentation and Classification of Computed Tomography Brain Images: An Approach Using One-Dimensional Kohonen Networks

This work is devoted to describe a potential use of the 1-Dimensional Kohonen Networks in the automatic non-supervised segmentation and classification of computed tomography brain slices. Possible perspectives of application include the automatic delineation of areas on the cerebral map and the automatic correlation between new clinical cases with previous boarded and closed cases. The classification is proposed in two phases. First, the images are segmented via a 1D Kohonen Network. One of the main aspects considered in this phase is related to the fact that tissue classification is achieved by taking in account the tissue and its associated neighborhood. By this way, it is possible to argue that the obtained tissue characterizations are sustained in the topology and geometry of the human cranium. The second phase is given by the classification of the whole set of segmented images via a second Kohonen Network. It is discussed how the final classes contain images which share specific properties. Index Terms Artificial Neural Networks, Kohonen Networks, Automatic Image Classification, Automatic Image Segmentation, Pattern Recognition.