Lester, Hava; (1999) Non-linear registration of medical images. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Non-linear_registration_of_med.pdf Download (23MB) | Preview

Abstract

This thesis provides a systematic analysis of registration algorithms for application to medical images. We divide our survey into four parts, each of which concentrates on a particular aspect of the algorithms. Linear methods are reviewed in respect to their selection of corresponding features, and their application to the inter-modality case. Non-linear methods are extensively analysed to understand the transition from a conceptual physical or statistical model to the mathematical model and its implementation. A chapter is devoted to hierarchical methods and their application to solving the local minima problem. Constructions of hierarchies are grouped as temporal variations in data complexity, in warp complexity and in model complexity. These divisions are paralleled in the classification of inhomogeneous methods, where the application of an algorithm varies spatially within the image. Thus we identify variances in relevance of data, in deformability and in chosen model type. In respect of these divisions, we have introduced a nomenclature to describe the restriction or otherwise of the deformation of selected regions in the image. We distinguish between passive and actively-deforming regions, between strongly and weakly deformable regions, and describe two specialisations of rigid regions, namely those which are motionless and those which are independently moving. The second main contribution of this work is in presenting three inhomogeneous variants to the viscous fluid registration algorithm, one for each of the three classes of inhomogeneity an algorithm may exhibit. In particular one of the variants exhibits a varying viscosity over the image. They are all tested for their ability to restrict the deformation of a specified region independently of the information contained within it. Finally we evaluate a selection of non-linear registration algorithms using both global and local registration metrics in a variety of tests. The dissertation concludes with three interesting suggestions for future projects.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as XMLCSVJSON

Archive Staff Only

View Item