Extensions of statistical shape models for medical imaging and computer vision / von M. E. Alma Eguizabal Aguado ; Erster Gutachter: Prof. Dr. Peter Schreier, Zweiter Gutachter: Prof. Dr.-Ing. Reinhold Häb-Umbach. Paderborn, 2020
Inhalt
- Abstract
- Zusammenfassung
- Acknowledgements
- Notation and Acronyms
- Contents
- I Introduction and background
- Introduction
- A framework for Statistical Shape Models
- An overview about shape models
- Motivation to extend Statistical Shape Models
- The need of robustness for fluoroscopic X-ray images
- Heuristic model selection
- Manual landmark registration
- Outlines and contributions of this thesis
- Statistical shape analysis
- Landmark-based shape
- Shape theory
- A shape manifold
- Shape invariant transformations
- A distance between shape observations
- Centroid and size
- A distance invariant to translation, scale and rotation
- Procrustes analysis
- Pre-shape and shape spaces
- Statistical Shape Models
- Statistical analysis of planar shapes
- A distance in the complex space
- Procrustes registration of two planar shapes
- Group-wise Procrustes registration
- Point Distribution Models
- Shape model fitting
- Statistical shape analysis with incomplete data
- II Developed work and contributions
- Robust Active Shape Models
- Motivation and preliminaries
- A weighting strategy for ASM
- GLS as a maximum likelihood problem
- Empirical determination of the residual errors
- Testing whether a target landmark is valid
- Results and discussion
- A model-order selection technique
- Motivation and preliminaries
- Source enumeration in array signal processing
- Information theory: entropy, differential entropy and mutual information
- Model selection
- The majorization-minimization optimization
- An information-theoretical approach
- Results and discussion
- Model-order selection in statistical shape models
- Procrustes registration of contours without correspondences
- Motivation and preliminaries
- Why to consider registration without correspondences
- Procrustes registration with correspondences
- Point set registration and the Iterative Closest Point algorithm
- Dynamic Programming and Dynamic Time Warping
- Dynamic Time Warping to establish correspondences
- Group-wise correspondence and registration
- DTW-based solution
- A probabilistic Procrustes registration
- Determining the weights
- Soft boundary condition
- Simultaneous pose and correspondences estimation
- Group-wise solution
- Results and discussion
- III Conclusions and future lines of research
- Conclusions and closing remarks
- Add robustness, keep simplicity
- The importance of the model order
- Registration without manual correspondences
- Contributions under development
- Procrustes registration of surfaces without correspondences
- Target landmark selection via sparse optimization
- Future Work
- IV Appendix
- Let there be Data
- Generation of a database of femoral shape
- Fluoroscopic images for computer-assisted surgery
- A Graphical User Interface to collect data
- Anatomical landmarks in the proximal femur
- Anatomical landmarks distal femur
- Correspondence and semi-landmarks
- Simulated shape data
- Other freely available databases
- Publications
- Lebenslauf
- List of Figures
- List of Algorithms
- Bibliography