Erik Bergvall presenterar sin licentiatavhandling Segmentation and Regularization in Velocity Encoded Cardiac Imaging
Torsdagen den 15 Juni 2006, kl 13.15 Lokal: MH:C
This thesis deals with the analysis of velocity encoded images of the heart, obtained by phase contrast MRI. Such data can be used to quantify regional cardiac motion, which is useful for diagnosis as well as evaluation of treatment.
Velocity data is used to construct non-linear spatiotemporal cardiac shape models as a tool for image segmentation. Cardiac shape is defined as a deformation of a template, defined by a diffeomorphic mapping, with constraints ensuring that the deformation if physically realizable. In the main part of the report, by means of a training set, physically realizable deformations are constructed as a chain of diffeomorphisms.
A second part deals with regularization of velocity data and motion estimates. Methods based on scale-space filtering, optical flow estimation and projection onto smooth splines are introduced. The performance of these methods is similar, but use of optical flow is less robust than the other methods.
A spline satisfying Navier’s equations for solid materials is derived, for means of representing elastic deformations. It is shown how this spline can be used to construct deformations that satisfy pointwise displacement constraints as well as how it can be used to approximate deformations defined on continuous domains.
The methods developed in this thesis have been applied to data from both healthy volunteers and patients with promising results.
Som granskare medverkar Docent Magnus Borga, Linköpings Tekniska Högskola
Gunnar Sparr, Matematikcentrum och Håkan Arheden, Klinisk fysiologi