Adaptive Verfahren zur automatischen Bildverbesserung kernspintomographischer Bilddaten als Vorverarbeitung zur Segmentierung und Klassifikation individueller 3D-Regionen des Gehirns

State-of-the-art digital image noise reduction does not have a procedure or algorithm combination with predefined parameters capable of reliably and successfully processing complex medical images of different image content, so that noise in the image is reduced while the anatomical structures are preserved. However, precisely this preprocessing step is required by segmentation, classification or visualization algorithms of medical images to allow greater flexibility when using such algorithms. Most of the currently used de-noising procedures depend on manual adjustment of the parameters. Only a few attempts were made to automate this adjustment, by defining the filter parameters according to the results of different estimation procedures. This thesis describes an iterative noise suppression procedure which can provide a fully automatic adjustment of the filter parameters by using a new optimization method. The filters used to process the medical images were anisotropic diffusion filters. These filters were designed to smooth images while preserving the edges between image regions. The evaluation method used determines image improvement indirectly by comparing the characteristics of the already suppressed noise to those from the acquisition system. This procedure can process 2D and 3D data sets and also offers the possibility of using multispectral information for better edge preservation. Synthetic and real MRI data were used to evaluate this automatic method. In a first step, T1-weighted averaged and mathematically generated T1-, T2- and PD-weighted software phantoms were used to test the main characteristics and performance of the procedure. The results were then compared to those of other commonly used non-linear filters. As a final step, samples from real data sets of different medical studies were processed and analyzed to demonstrate the advantage of this automatic procedure in practical applications