| Popis: |
In the DBS surgical treatment the location of the electrodes has direct influence on the success rate of treatment. Considering the fact that MRI imaging is not compatible with metal, post-operative CT images should be used to locate the electrodes on pre-operative MRI images which contain information about soft tissues. It is impossible to locate them directly since the area around the electrodes on the CT image has been corrupted due to the blockage of x-ray by them. This work attempts to precisely estimate the position of the DBS electrodes inside the brain by means of images gathered from the CT-scanner and MRI device. In order to do this, all types of possible CT artifacts and the physics behind has been carefully studied and a state of the art research on MAR techniques has been performed. Apart from hardware MAR techniques and also methods which are applicable only by having the model of the gantry (Cylindrical scanner assembly) and raw data, after the formation of the image it is not possible to completely recover all lost intensity levels without exposing the result to the risk of secondary artifact as well as electrode dislocation which might be minor yet critical for this application. Considering the latter, a novel technique named junction method developed. Instead of eliminating the artifacts, the method takes advantage of them, considering the fact that they are signatures of the electrodes. To achieve this, first the brain extracted from the whole image by defining a brain mask. Later the edges are intensified by applying a Gaussian convolution followed by a second measure of the second derivative of the image along all directions. Next all lines are detected by the Hough transform and after filterations the intersections of interest are specified. In the second part of this project, state of the art research on multi-modality medical image registration techniques was carried out. Advantages and disadvantages of each method in relation to this specific case were studied. Registration technique selection could not be done without investigation of all types of discrepancies which might exist between two image modalities; therefore discrepancies are being discussed. Nonrigid Affine registration algorithm suggested along with logarithmic contrast enhancement and Gaussian smoothening as preprocessing steps, based on experiences performed on an available image set. The mutual information, an information-theoretic criterion used as the measure of registration with the help of calculation of a joint histogram of the two modalities. All the algorithms starting from CT images to MRI images are implemented in Matlab software and the resulting image shows a close matching. |
