Graph based MR image registration

In this thesis, graph-based medical image registration methods were reviewed. Additionally a new similarity measure based on the joint saliency map weighted orthogonal regression on the graph is proposed. The proposed similarity measure is expressed as the sum of the joint saliency map weighted values with ortogonal residuals (errors) over the generated entropic regression line. Many comprehensive comparisons are made for the validation of his measure. While making medical image registration, the used graph structures were examined in three different ways: feature selection, similarity measure and transformation parameters optimization. In feature selection, while creating entropic graph, minimal points expressing the image were located. In similarity measure criteria, in image registration, how much reference and test image were resembled to each other with the created entropic graphs are measured. Finally, in optimization selection criteria, in order to register reference and test images, methods for finding the transformation have been searched and the most appropriate ones are defined.