Non–linear Registration of Pre– and Intraoperative Volume Data Based On Piecewise Linear Transformations

For the planning of minimal invasive brain surgery detailed knowledge of the individual anatomical structures is necessary. Therefore in medical practice high–resolution MR image data is recorded preoperatively. Due to tissue resection and cerebrospinal fluid leakage both shape and position of the brain change during the intervention (brain–shift). Therefore intraoperative data is used in addition, providing exact anatomical information. In order to accommodate the deformations of tissue for computer assisted surgery, nonlinear registration of the voxel data must be performed. In this context we propose a new voxel–based approach based on maximization of mutual information. For fast evaluation the nonlinear deformation is modeled by adaptively subdividing the data set into piecewise linear patches. The parameters of this multidimensional transformation are optimized using Powell’s direction set method. Since 3D– texture hardware is exploited to evaluate trilinear interpolations, the registration process is significantly accelerated, which is of vital importance for the application in medical context.