|M.Sc Student||Zaideman Oded|
|Subject||3D Remeshing of Bone Micro Structures from Micro-CT/MRI|
Images Using Grid-Based and Neural Network
|Department||Department of Mechanical Engineering||Supervisor||Professor Anath Fischer|
|Full Thesis text|
Recently, Reverse Engineering (RE) has become a major field in medical imaging. RE refers to the process of reconstructing a computerized model from a digitized object, usually derived from a laser scanner for surface models and from CT/MRI images for volumetric models. Commercial RE modules often fail to analyze the scanned data with the high accuracy required to preserve topology, especially for micro-structures. Current reconstructed mesh models are not suitable for mechanical analysis due to their irregular elements. Moreover, they are often noisy and inaccurate. Hence, the data must be re-meshed.
This research focuses on the problem of remeshing a 3D computerized mesh model of a bone, scanned by mCT technology. This research propose dividing the problem into three phases: defining sub-meshes in a grid-based structure, re-meshing each sub-mesh using the Neural Network method, and merging the sub-meshes into a global mesh. The advantages of this method are examined and its performance is analyzed.
In this research, the proposed grid-based neural network method and its implementation are described. A neural network method was used to reconstruct high quality meshes from large-scale clouds of points. The method was then applied to bone micro-structures, where it proved to be highly time consuming. Therefore, a parallel approach with a grid-based structure has been implemented to cope with micro-structures. By partitioning the complex micro-structure into grid cells, we achieve in each cell a simpler structure, geometrically and topologically. The method is demonstrated on real bone micro-structures from the medical field.