Artikel
Application of the variational non-rigid JIRDED (Joint Image Registration, Denoising and Edge Detection) method on 3D-MRI and 2D-photo to 3D-MRI registration
3D-MRT und 2D-Photo-Registrierung mit der JIRDED (Joint Image Registration, Denoising and Edge Detection) Methode
Suche in Medline nach
Autoren
Veröffentlicht: | 11. April 2007 |
---|
Gliederung
Text
Objective: 3D-MR Image registration nowadays is a routine technology in neurosurgical planning and intraoperative neuronavigation, especially in eloquent tumour location or in epilepsy surgery preoperative invasive electrophysiological examination is necessary to pinpoint different brain functions. With the Joint Image Registration, Denoising and Edge Detection (JIRDED) Method we developed a reliable algorithm to detect the image edges, denoise the images and symmetrically transform them for non-rigid image registration simultaneously.
Methods: The algorithm was performed on inter-object 3D-MRI-to-MRI and 3D-MRI T1-to-T2 weighted data sets. High-resolution 3D-T1-MR imaging was used. Multi-resolution procedures with 10 iterations in all levels respectively and visual control of the results with regard to precise alignment of the volume shape, contours, ventricular system and phase field functions were accomplished. Additionally 2D-photo-to-3D-brain-surface asymmetric registrations were performed. A digital camera was used to take pictures of the exposed cortex with the fitted electrode. Volume rendering of the brain surface was accomplished with MRIcro software. Again, visual control of the results with regard to alignment of visible gyri and sulci was performed. The correct matchings of the data were visualized by a so-called “interlace-stripe”, showing the corresponding volumes before and after registration.
Results: The algorithm showed allows far reliable edge detection of registered images. This was ensured in a spatial multi-scale implementation scheme. All MRI data sets showed satisfactory one-to-one contour matching after registration of both intra-object and inter-object unimodal 3D-MRI. In unimodal intra-object registration T1- and T2-weighted MRI of the same individual with different intensities due to different tissue properties precise alignment of edges and geometry was achieved. The non-rigid algorithm worked accurately also on inter-object 3D-MRI of different individuals. Also the multimodal application of 2D-photo-to-3D-brain-surface registration results in nearly perfect matching and solved the problem of geometric ambiguity between cortical vessels and contours of gyri and sulci.
Conclusions: The JIRDED algorithm simultaneously accomplishes edge detection, image denoising and consistently registers image data using the detected contour features. The algorithm works precisely on intra- and inter-object 3D data sets and on multimodal 2D-photo-to-3D-MRI registration. The algorithm shows future potential with regard to precise image alignment in neuronavigational applications.