Advanced

Vessel segmentation for time resolved phase contrast MRI

Bidhult, Sebastian LU (2013) In Master's Theses in Mathematical Sciences FMA820 20122
Mathematics (Faculty of Technology) and Numerical Analysis
Abstract
Quantication of cardiovascular flow and blood volumes are useful tools in diagnosing cardiovascular disease such as congenital heart defects and different kinds of valve leakage. Medical imaging techniques enables non-invasive analysis of anatomy and physiology. In order to perform flow quantification from medical image sequences, the boundaries of the
vessels of interest are usually delineated manually by medical professionals. This is a time consuming process and the result depends, to a high degree, on user experience. This thesis presents an automated vessel segmentation method for the main vessels around the
heart from velocity encoded Magnetic Resonance Imaging sequences. The proposed method only require a manual delineation in one... (More)
Quantication of cardiovascular flow and blood volumes are useful tools in diagnosing cardiovascular disease such as congenital heart defects and different kinds of valve leakage. Medical imaging techniques enables non-invasive analysis of anatomy and physiology. In order to perform flow quantification from medical image sequences, the boundaries of the
vessels of interest are usually delineated manually by medical professionals. This is a time consuming process and the result depends, to a high degree, on user experience. This thesis presents an automated vessel segmentation method for the main vessels around the
heart from velocity encoded Magnetic Resonance Imaging sequences. The proposed method only require a manual delineation in one image. The algorithm is based on an active contour, using the Euler-Lagrange equation together with internal and external forces designed from
a set of fundamental assumptions regarding vessel shape and behaviour. More specically, constraints were applied to the geometrical shape and elasticity of a vessel. Validation of the method was performed by comparing the detected stroke volume with manual delineation, and also by measuring the segmentation overlapping of the two methods. On a test set of 20 patients, 19 resulted in excellent segmentation agreement with manual delineations, with a mean Dice coefficient over 0.8. However, performance instability was observed when changing the values of two algorithm parameters, and one of the patients in the set constantly
resulted in segmentation failure for all tested parameter combinations. The relative variability in stroke volume between the proposed algorithm and manual delineations was, at best, 6 +- 3.6%. This is comparable to the interobserver variability from a previous physiological
study 1 of 3 +- 4%, which indicates the potential of the suggested method if improvements in robustness and stability is implemented. (Less)
Please use this url to cite or link to this publication:
author
Bidhult, Sebastian LU
supervisor
organization
alternative title
Kärlsegmentering i tidsupplösta MRI sekvenser
course
FMA820 20122
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Medical image analysis, Active contours, Euler Lagrange equations, Expectation Maximization algorithm.
publication/series
Master's Theses in Mathematical Sciences
report number
LUTFMA-3243-2013
ISSN
1404-6342
other publication id
E12
language
English
id
3802951
date added to LUP
2013-09-20 12:16:59
date last changed
2013-09-20 12:16:59
@misc{3802951,
  abstract     = {Quantication of cardiovascular flow and blood volumes are useful tools in diagnosing cardiovascular disease such as congenital heart defects and different kinds of valve leakage. Medical imaging techniques enables non-invasive analysis of anatomy and physiology. In order to perform flow quantification from medical image sequences, the boundaries of the
vessels of interest are usually delineated manually by medical professionals. This is a time consuming process and the result depends, to a high degree, on user experience. This thesis presents an automated vessel segmentation method for the main vessels around the
heart from velocity encoded Magnetic Resonance Imaging sequences. The proposed method only require a manual delineation in one image. The algorithm is based on an active contour, using the Euler-Lagrange equation together with internal and external forces designed from
a set of fundamental assumptions regarding vessel shape and behaviour. More specically, constraints were applied to the geometrical shape and elasticity of a vessel. Validation of the method was performed by comparing the detected stroke volume with manual delineation, and also by measuring the segmentation overlapping of the two methods. On a test set of 20 patients, 19 resulted in excellent segmentation agreement with manual delineations, with a mean Dice coefficient over 0.8. However, performance instability was observed when changing the values of two algorithm parameters, and one of the patients in the set constantly
resulted in segmentation failure for all tested parameter combinations. The relative variability in stroke volume between the proposed algorithm and manual delineations was, at best, 6 +- 3.6%. This is comparable to the interobserver variability from a previous physiological
study 1 of 3 +- 4%, which indicates the potential of the suggested method if improvements in robustness and stability is implemented.},
  author       = {Bidhult, Sebastian},
  issn         = {1404-6342},
  keyword      = {Medical image analysis,Active contours,Euler Lagrange equations,Expectation Maximization algorithm.},
  language     = {eng},
  note         = {Student Paper},
  series       = {Master's Theses in Mathematical Sciences},
  title        = {Vessel segmentation for time resolved phase contrast MRI},
  year         = {2013},
}