A knowledge-based deformable surface model for analysis of medical images
Medical images are challenging for segmentation. Deformable models proved to be one of the most effective tools for this purpose, however they still have several shortcomings and face several problems when used for images with objects that have low contrast, multiple edges (of other objects) in thei...
Gespeichert in:
| 1. Verfasser: | |
|---|---|
| Format: | Dissertation |
| Sprache: | Englisch |
| Veröffentlicht: |
ProQuest Dissertations & Theses
01.01.2001
|
| Schlagworte: | |
| ISBN: | 9780493278612, 0493278613 |
| Online-Zugang: | Volltext |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Zusammenfassung: | Medical images are challenging for segmentation. Deformable models proved to be one of the most effective tools for this purpose, however they still have several shortcomings and face several problems when used for images with objects that have low contrast, multiple edges (of other objects) in their vicinity, or do not have a continuous boundary in the image. They also suffer in general from problems such as initialization, self-cutting, and concave shape. In this research, we developed and implemented a two-dimensional (2D) and three-dimensional (3D) deformable model for analysis of medical images. We used the 2D deformable model to segment structures with low contrast and multiple or discontinuous edges. To do this, we developed an edge-tracking algorithm for external forces, adaptive force weights, and interpolation of energies. We further developed a 2D warping model that uses this deformable contour to extract the boundaries and create a mesh-map in each image using equidistance contours. Thin-plate splines or inverse-distance weights are used for interpolation between the mesh-map grids. We developed a 3D deformable surface model as the extension of the deformable contour to 3D. The model has a closed discrete structure based on polygonal facets. We developed a method based on LSE estimation of Dupin indicatrix to calculate internal forces and use gradient-based operator for external forces. We remedy self-cutting by extracting principal axis and performing reslicing followed by triangulation of the model. We propose a method for creating the initial surface from the individual 2D contours, and use multi-resolution and resampling of the surface and re-triangulation to further improve the model deformation. The model is initialized using a rule-based expert system. We extended the 2D warping model to 3D. The developed models have been used in several challenging applications among which, segmentation of hippocampus from brain MRI and prostate from ultrasound images, and the warping have been extensively investigated. Comparison to manual results shows excellent model performance. |
|---|---|
| Bibliographie: | SourceType-Dissertations & Theses-1 ObjectType-Dissertation/Thesis-1 content type line 12 |
| ISBN: | 9780493278612 0493278613 |