In this dissertation a new approach for non-rigid medical im-udage registration is presented. It relies onto a probabilistic frameworkudbased on the novel concept of Fuzzy Kernel Regression. The theoricudframework, after a formal introduction is applied to develop severaludcomplete registration systems, two of them are interactive and oneudis fully automatic. They all use the composition of local deforma-udtions to achieve the final alignment. Automatic one is based onto theudmaximization of mutual information to produce local affine aligmentsudwhich are merged into the global transformation. Mutual Informationudmaximization procedure uses gradient descent method. Due to theudhuge amount of data associated to medical images, a multi-resolutionudtopology is embodied, reducing processing time. The distance basedudinterpolation scheme injected facilitates the similairity measure op-udtimization by attenuating the presence of local maxima in the func-udtional. System blocks are implemented on GPGPUs allowing efficientudparallel computation of large 3d datasets using SIMT execution. Dueudto the flexibility of Mutual Information, it can be applied to multi-udmodality image scans (MRI, CT, PET, etc.).udBoth quantitative and qualitative experiments show promising resultsudand great potential for future extension.udFinally the framework flexibility is shown by means of its succesfuludapplication to the image retargeting issue, methods and results areudpresented.
展开▼
