|Daphne Yu's Research Page||[Résumé]||[Links]|
Accurate reconstruction of the human cerebral cortex from magnetic resonance images (MRI) is important for brain morphometric analysis, image-guided surgery, and functional mapping. Our lab has been working on a cortical surface reconstruction method that employs fuzzy segmentation, isosurfaces and deformable surface models. Unfortunately, global quantitative validation of the cortical surface model is not feasible due to the lack of a true and continuous representation of the cortical surface. My main research project is to validate the deformable surface model using a metasphere computational phantom instead. A metasphere is a mathematically defined 3-D continuous object with a star-shaped surface that has convolutions similar to the cortex. This should serve as a starting point for providing some measure of accuracy of the deformable model in relation to the properties of the modeled object and data quality.
This illustrates (Top row) one slice of the simulated white matter and gray matter image volumes from a metasphere phantom with 7 convolutions, and (Bottom row) one slice of the white matter and gray matter image volumes segmented from a brain. The metasphere's shape is quite simple, but still captures the concave nature (like the brain) which often poses problems for deformable models.
This shows a very bumpy metasphere (left) and a smooth metaphsere (right) surface, with the error of the deformable models mapped onto them. Red regions indicates higher error.
This is an interactive software, built based on the VTK (Visualization Tookit) & Tcl/Tk, that I'm currently working on. Editing and visualizing complex 3-D structure such as cortical surfaces can be very difficult. This tool is intended to provide a way for a user to manipulate these 3-D surface by interacting with the 2-D intersection between the 3-D surface and the image plane from which the surface is reconstructed. More information about this project can be found here.
This is just a Tcl/Tk GUI that I worked on a year ago or so. The idea is to provide an semi-automated pipeline for data processing such that massive data processing of brain surfaces can be a bit less painful.