Reconstruction and analysis of soft-tissue organs and tubular structures using three-dimensional statistical deformable models
Student thesis: Doctoral Thesis
Related Research Unit(s)
Deformable models have aroused much interest and found various applications in the fields of computer vision and medical imaging. Particularly, they provide an extensible framework to segment and reconstruct shapes from medical and biomedical dataset. More recently, statistical deformable model opens up a promising direction for encoding and learning “tailored” models from training sample set in a stochastic way. In this thesis, we propose two novel statistical models and the associated techniques for the reconstruction of soft-tissue organs such livers and tubular structures such as vasculatures. To address the small sample size problem that frequently encountered in biomedical applications, we define a Multi-resolution Integrated Model for Soft-Tissue organs (MISTO) and a Statistical Assembled Model for Tubular Structures (SAMTUS). Both of the models are constructed in a hierarchical way to represent the most significant deformations from the training set as well as to generate representative variation modes of the organ shapes. The clutter surrounding of the surface points are formulated in terms of the external functional which is also learnt from the training samples. By combining the powerful shape models and context constraints, the object segmentation and reconstruction process can be carried out very effectively. To avoid the local minimum during model optimization, the deformation strategies are designed in the way that moves the reliable parts of the surface prior to the unsure parts. The experimental and validation results verify that our proposed approaches can be successfully and robustly applied to the reconstruction of the soft-tissue organs like the human liver and the tubular structures like the vasculature of the zebrafish embryo. The major contributions of our approaches are that we extend the traditional point distribution model to address open problems associated with reconstructing significantly deformable 3-D anatomies in cluttered surrounding, and we propose effective ways to formulate the perceptual knowledge of the anatomies and make use of it in the correspondence matching, model constructing and surface deformation.
- Tissues, Diagnostic imaging, Blood-vessels, Digital techniques, Imaging