Use this URL to cite or link to this record in EThOS:
Title: A hybrid similarity measure framework for multimodal medical image registration
Author: Reel, Parminder
ISNI:       0000 0004 5989 335X
Awarding Body: Open University
Current Institution: Open University
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Access from Institution:
Medical imaging is widely used today to facilitate both disease diagnosis and treatment planning practice, with a key prerequisite being the systematic process of medical image registration (MIR) to align either mono or multimodal images of different anatomical parts of the human body. MIR utilises a similarity measure (SM) to quantify the level of spatial alignment and is particularly demanding due to the presence of inherent modality characteristics like intensity non-uniformities (INU) in magnetic resonance images and large homogeneous non-vascular regions in retinal images. While various intensity and feature-based SMs exist for MIR, mutual information (MI) has become established because of its computational efficiency and ability to register multimodal images. It is however, very sensitive to interpolation artefacts in the presence of INU with noise and can be compromised when overlapping areas are small. Recently MI-based hybrid variants which combine regional features with intensity have emerged, though these incur high dimensionality and large computational overheads. To address these challenges and secure accurate, efficient and robust registration of images containing high INU, noise and large homogeneous regions, this thesis presents a new hybrid SM framework for 2D multimodal rigid MIR. The framework consistently provides superior quantitative and qualitative performance, while offering a uniquely flexible design trade-off between registration accuracy and computational time. It makes three significant technical contributions to the field: i) An expectation maximisation-based principal component analysis with mutual information (EMPCA-MI) framework incorporating neighbourhood feature information; ii) Two innovative enhancements to reduce information redundancy and improve MI computational efficiency; and iii) an adaptive algorithm to select the most significant principal components for feature selection. The thesis findings conclusively confirm the hybrid SM framework offers an accurate and robust 2D registration solution for challenging multimodal medical imaging datasets, while its inherent flexibility means it can also be extended to the 3D registration domain.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral