Surface mismatch inverse based methods in brain deformation modeling

An accurate registration between the preoperative images and the current surgical scene is critical to image-guided neurosurgery. However, during a neuro-surgical intervention, brain tissues shift and degrade the accuracy of the registration. Gravity can cause the brain to shift up to 10 mm. Some surgical procedures like tumor retraction and tumor resection can significantly compromise the accuracy of preoperative-based image guidance. In order to keep an accurate tracking of surgical instruments, brain deformation needs to be estimated and the preoperative images need to be updated to reflect the current status of the brain during surgery. Incorporating intra-operative data with a biomechanical model is able to compensate for brain deformation. Three major improvements in the model have been developed in this thesis to improve the computational efficiency of the model based non-rigid registration method. First, the adjoint equations are solved iteratively to incorporate more measurements into the model and to improve the speed of computation. Second, a surface-based model-data mismatch method has been developed to replace the current point-based misfit approach to make the solving process more robust. In this method, the volume between the measured surface and model predicted surface is minimized. Third, a parameterized surface method is investigated. The measured data and the corresponding model computed data are fitted to parameterized spherical surfaces by applying a least squares fitting method. The model-data misfits of data points are then replaced by the difference between the two sets of parameters in the iterative adjoint equation algorithm. This method aims at eliminating the steps of segmenting brain structures and generating displacement vectors, so the modeling process can be more robust. The first two inversion schemes have been applied to clinical procedures and the computational speed was improved. The third method has been tested with 2D and 3D simulations, the results show the model captured deformation up to 79.3%.Brain ventricular deformation was also studied through five in vivo feline experiments. The ventricles were modeled as cavities with appropriate boundary conditions applied and the computational model predicted the ventricular deformation. The measured displacement data which were extracted from pre-drainage and post-drainage magnetic resonance (MR) images were incorporated into the model through the iterative AEM method. The results indicate that the computational modeling of the brain and ventricular system captured 33% of the ventricle deformation on average and the model estimated intraventricular pressure was accurate in the range of 90% compared to the recorded pressure during the hydrocephalus experiments.