Author(s):
Libório, André Filipe Cidra Duarte Rocha 
Date: 2011
Persistent ID: http://hdl.handle.net/10316/17592
Origin: Estudo Geral - Universidade de Coimbra
Subject(s): Imagem biomédica; Tomografia computorizada
Document details
Development of co-registration techniques for comparison of whole-body PET/CT s...
Description
The objective of this thesis is to develop an algorithm that co-registers, in a single volume, whole-body Human PET/CT Scans taken at different instants in time, to facilitate the comparison of studies in situations where the patient had to multiple PET/CT exams. The idea is to isolate, from the initials CT volumes, the rigid body parts that can be transformed with a rigid transformation – the bone structures. Bone structures were obtained with segmentation (the process of dividing a volume into subsections, here based on intensity values) and a further labeling, taking into account the connectivity between neighboring voxels. Once the bone structures were labeled in both volumes, the next step was the establishment of its correspondences between the regions in the two volumes. This was based on the number of voxels associated with each region (size of region) and with the Euclidian distance between the centers of mass of regions from different volumes. Once this correspondence was achieved, the rigid transformation model is constructed and applied to each regions of the volume, which was defined as the Target volume. The results of the developed method are still far from those required for clinical applications. We hope that, with future developments of the presented algorithm, will lead to the improvement of medical image analysis in this field
The objective of this thesis is to develop an algorithm that co-registers, in a single volume, whole-body Human PET/CT Scans taken at different instants in time, to facilitate the comparison of studies in situations where the patient had to multiple PET/CT exams. The idea is to isolate, from the initials CT volumes, the rigid body parts that can be transformed with a rigid transformation – the bone structures. Bone structures were obtained with segmentation (the process of dividing a volume into subsections, here based on intensity values) and a further labeling, taking into account the connectivity between neighboring voxels. Once the bone structures were labeled in both volumes, the next step was the establishment of its correspondences between the regions in the two volumes. This was based on the number of voxels associated with each region (size of region) and with the Euclidian distance between the centers of mass of regions from different volumes. Once this correspondence was achieved, the rigid transformation model is constructed and applied to each regions of the volume, which was defined as the Target volume. The results of the developed method are still far from those required for clinical applications. We hope that, with future developments of the presented algorithm, will lead to the improvement of medical image analysis in this field
Document Type
Master Thesis
Language English
Advisor(s) Ferreira, Nuno Chichorro; Caramelo, Francisco Amado
Language English
Advisor(s) Ferreira, Nuno Chichorro; Caramelo, Francisco Amado
| Financiadores do RCAAP | |||||||
|
|
|
|
|
|
||