Evaluating non-rigid image registration performance is a difficult problem since there is rarely a “gold standard” (i.e., ground truth) correspondence between two images. The Non-rigid Image Registration Evaluation Project (NIREP) was started to develop a standardized set of common databases, evaluation statistics and a software tool for performance evaluation of non-rigid image registration algorithms. The goal of the work in this thesis is to build up common image databases for rigorous testing of non-rigid image registration algorithms, and compare their performance by a diverse set of evaluation statistics on our multiple well documented image databases. The well documented databases as well as new evaluation statistics have been and will be released to public research community. The performance of five non-rigid registration algorithms (Affine, AIR, Demons, SLE and SICLE) was evaluated using 22 images from two NIREP evaluation databases. Six evaluation statistics (Relative Overlap, Intensity Variance, Normalized ROI overlap, alignment of calcarine sulci, Inverse Consistency Error and Transitivity Error) were used to evaluate and compare registration performance. This thesis provides a complete and accurate reporting of evaluation tests so that others are able to get access to these results and make a comparison of registration algorithms they concerned in their specific use. Moreover, this work followed the recommendations of the Standards for Reporting of Diagnostic Accuracy (STARD) initiative to disclose all relevant information for each non-rigid registration validation test.

Advances of magnetic resonance imaging (MRI) techniques enable visualguidance to identify the anatomical target of interest during the image guidedintervention(IGI). Non-rigid image registration is one of the crucial techniques,aligning the target tissue with the MRI preoperative image volumes. As thegrowing demand for the real-time interaction in IGI, time used for intraoperativeregistration is increasingly important. This work implements 3D diffeomorphicdemons algorithm on Nvidia GeForce GTX 1070 GPU in C++ based on CUDA8.0.61 programming environment, using which the average registration time hasaccelerated to 5s. We have also extensively evaluated GPU accelerated 3D diffeomorphicregistration against both CPU implementation and Matlab codes, and theresults show that GPU implementation performs a much better algorithm efficiency.

La fusion d’images TDM de phase artérielles et veineuses est cruciale afin d’assurer une meilleure planification chirurgicale. Cependant, le recalage non-rigide d’images abdominales est encore un challenge à cause de la respiration qui fait glisser les viscères abdominaux le long de la paroi abdominale, créant ainsi un champ de déformation discontinu. L’objectif de cette thèse est de fournir un outil de recalage précis pour les images TDM multi-phases de l’abdomen.Comme la zone de glissement dans l’abdomen est difficile à segmenter, nous avons d’abord implémenté deux outils de segmentation interactifs permettant une délinéation en 10 minutes de la paroi abdominale et du diaphragme. Pour intégrer ces zones de glissement comme a priori, nous réalisons le recalage sur de nouvelles images dans lesquelles la paroi abdominale et les viscères thoraciques ont été enlevés. Les évaluations sur des données de patient ont montré que notre approche fournit une précision d’environ 1 mm. / The fusion of arterial and venous phase CT images of the entire abdominal viscera is critical for a better diagnosis, surgi-cal planning and treatment, since these two phase images contain complementary information. However, non-rigid regis-tration of abdominal images is still a big challenge due to the breathing motion, which causes sliding motion between the abdominal viscera and the abdo-thoracic wall. The purpose of this thesis is to provide an accurate registration method for abdominal viscera between venous and arterial phase CT images.In order to remove the sliding motion effect, we decide to separate the image into big motion and less motion regions, and perform the registration on new images where abdo-thoracic wall and thoracic viscera are removed. The segmentation of these sliding interfaces is completed with our fast interactive tools within 10 minitues. Two state-of-the-art non-rigid registration algorithms are then applied on these new images and compared to registration obtained with original images. The evaluation using four abdominal organs (liver, kidney, spleen) and several vessel bifurcations shows that our approach provides a much higher accuracy within 1 mm.

Cílem předkládané diplomové práce je pomocí lícování obrazů přesně popsat distorzní pole pro následné odstranění geometrické distorze. Snímky zkreslené geometrickou distorzí pochází z prozařovacího elektronového mikroskopu. První část práce se zabývá zejména teorií spojenou s elektronovou mikroskopií, vznikem geometrické distorze a samotnou obrazovou registrací s důrazem na intenzitní flexibilní metody lícování. Ve druhé části je pak představena vytvořená metoda pro modelování geometrické distorze a lícování obrazů postižených slabou geometrickou distorzí. Vyvinutá metoda je následně otestována na testovacích i reálných datech a srovnána s existujícími popsanými metodami pro obrazovou registraci (například open-source softwarem Elastix).