Techniques for Deformable Organ Registration in DCE Framework

In the last decades, the development of non-invasive medical imaging modalities enabled remarkable progress in clinical diagnostics. Many diseases are nowadays detected by medical imaging techniques, which allows for earlier diagnosis and therapeutic intervention with a higher effi ciency. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is a clinical technique devoted to the study of lesions in organs. It is based on the analysis of perfusion dynamics of a contrast agent (CA) in the examined tissues by acquiring an image series over time, before, during and after the arrival of the CA. The acquired signal is analyzed for a selected region of interest (ROI) or individual pixels in each 2D slice of the series to generate an enhancement curve (EC), which re ects the tissue's response to the arrival of the CA by time-dependent enhancement valu of the detected intensity. These curves provide a set of parameters, commonly visualized by 2D color maps, that yield valuable information for diagnosis. However, for a correct diagnosis, it is crucial to determine the EC with a high accuracy, which is often compromised by artifacts within the acquired time series due to patient motion. DCE algorithms assume that the analyzed ROI does not modify its shape and position over the entire time series, which is usually not a valid assumption, since the examined non-rigid structures are ffected by patient motion and can change in position, shape and brightness. The compensation for such artifacts is complex, since the detected signal intensities intrinsically vary over time due to the diffusion of the CA. In this thesis, we present a novel technique to compensate artifacts in DCE induced by motion/deformation of the ROI. The algorithm uses deformable active contours (AC), also called snakes, to defi ne the deformation of the ROI in each image. We optimized the standard AC method by adding an additional term to the energy functional that is minimized during the procedure. Speci cally, we included a distance map calculated with the Chamfer transform into e external forces that are responsible for the attraction of the snake to the edges in an image. This results in a considerably more accurate detection of the contour of the ROI by the snake and thus in a better segmentation. Subsequently, the center of mass and the principal axes for each frame are determined and used for a rototranslation compensation of misalignment between the frames. The algorithm was applied to different organs of the abdominal part and the results clearly show an improved accuracy of the corrected EC and the resulting parameters, when compared to the uncorrected EC. In addition to inter-frame misalignments, motion artifacts also cause displacements in the transverse direction relative to the image plane, thereby generating inconsistencies in the shape of the organ pro les in the same temporal series. Therefore, we extended our algorithm to four dimensions by using a hybrid approach based on ACs and a template matching algorithm based on the Chamfer distance. The frame-by-frame method segments the ROI of a particular frame via the AC model, subsequently, the fi nal contour of the time frame is compared to the images of the same and adjacent planes of the successive frame and the best match is assigned to the previous image. Finally, we introduce a fast and user-friendly tool to analyze 4D DCE-MRI data. The results can be visualized in different 3D representations and provide quantitative volumetric information in order to improve the understanding of the contrast agent perfusion in pathological tissues and the resulting diagnosis.