Use this URL to cite or link to this record in EThOS:
Title: Automatic whole heart segmentation based on image registration
Author: Zhuang, X.
ISNI:       0000 0004 2728 3311
Awarding Body: University College London (University of London)
Current Institution: University College London (University of London)
Date of Award: 2010
Availability of Full Text:
Access from EThOS:
Access from Institution:
Whole heart segmentation can provide important morphological information of the heart, potentially enabling the development of new clinical applications and the planning and guidance of cardiac interventional procedures. This information can be extracted from medical images, such as these of magnetic resonance imaging (MRI), which is becoming a routine modality for the determination of cardiac morphology. Since manual delineation is labour intensive and subject to observer variation, it is highly desirable to develop an automatic method. However, automating the process is complicated by the large shape variation of the heart and limited quality of the data. The aim of this work is to develop an automatic and robust segmentation framework from cardiac MRI while overcoming these difficulties. The main challenge of this segmentation is initialisation of the substructures and inclusion of shape constraints. We propose the locally affine registration method (LARM) and the freeform deformations with adaptive control point status to tackle the challenge. They are applied to the atlas propagation based segmentation framework, where the multi-stage scheme is used to hierarchically increase the degree of freedom. In this segmentation framework, it is also needed to compute the inverse transformation for the LARM registration. Therefore, we propose a generic method, using Dynamic Resampling And distance Weighted interpolation (DRAW), for inverting dense displacements. The segmentation framework is validated on a clinical dataset which includes nine pathologies. To further improve the nonrigid registration against local intensity distortions in the images, we propose a generalised spatial information encoding scheme and the spatial information encoded mutual information (SIEMI) registration. SIEMI registration is applied to the segmentation framework to improve the accuracy. Furthermore, to demonstrate the general applicability of SIEMI registration, we apply it to the registration of cardiac MRI, brain MRI, and the contrast enhanced MRI of the liver. SIEMI registration is shown to perform well and achieve significantly better accuracy compared to the registration using normalised mutual information.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available