National Technical University of Athens (NTUA)
Τμηματοποίηση, ευθυγράμμιση και αυτόματη ταξινόμηση ιατρικών δεδομένων μίκρο-τομογραφίας σπογγώδους οστού
Abstract
dc:descriptionThe main goal of this PhD thesis, is the processing of micro-CT trabecular bone data using advanced computational algorithms. In the following dissertation, three novel automatic methodologies are presented regarding the segmentation and registration of trabecular bone micro-CT images, and the classification of data from the extraction of morphometrical features from the aforementioned images. Initially, an introduction to the principal elements and dynamic processes of bone is made, along with a description of the main degenerative diseases, and analytic and imaging methodologies. Then, the image segmentation system is presented, following a study of the advantages and disadvantages of using typical image segmentation algorithms on trabecular bone micro-CT images. The image segmentation system is implemented using a novel segmen-tation framework under the name Independent Active Contours Segmentation that combines a set of novel segmentation techniques based on Active Contours. The proposed methodology enables the robust wide application of Active Contours techniques on full trabecular bone datasets, while significantly enhancing the segmentation system’s performance on Phantom and real data. At the same time, the proposed method is applied on other medical image modalities, such as Computed Tomography and Histology images, showing the capability of a wider application. Moving on, the micro-CT image registration system is presented and is applied on samples’, pre and post failure datasets, aiming at developing a quantitative criterion for the identification of fracture zone. Within this research 2D and 3D registration techniques are presented, followed by some results using appropriate metrics in order to evaluate their accuracy. Moreover, an automatic classification system for the identification of fracture zone from the values of the various morphometrical parameters extracted from the segmented images’ Volumes of Interest is presented. This study combines Ensemble Classification and Imbalanced Learning techniques with well-established classification algorithms, aiming at finding the best possible combination. The final system is tested on a large dataset were it provides very high performance compared to the simpler combinations, without significantly increasing the complexity and the computational burden of the whole procedure. Finally, an evaluation of the most important morphometrical parameters takes place using Feature Selection techniques in Imbalanced Learning datasets, aiming at the extraction of possible biomarkers for the fracture zone.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2016
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Korfiatis, Vasileios
- Κορφιάτης, Βασίλειος
Subjects
dc:subject × 18- Μίκρο-τομογραφία
- Ψηφιακή επεξεργασία εικόνας
- Εικόνες, τμηματοποίηση
- Ευθυγράμμιση ιατρικών δεδομένων
- Μηχανική μάθηση
- Σπογγώδες οστούν
- Micro-CT
- Digital image processing
- Image segmentation
- Medical image registration
- Machine learning
- Trabecular bone
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Βιοϊατρική Μηχανική
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
- Medical Engineering
Rights
dc:rights- Statement dc:rights
-
- BY_NC_SA
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/38557