Advanced

Semi-Automatic Segmentation of the Right Ventricle in MR Images

Gleisner, Gabriella LU (2016) In Master's Theses in Mathematical Sciences FMA820 20162
Mathematics (Faculty of Engineering)
Abstract
Cardiac magnetic resonance is widely used to non-invasive image the heart, for both clinical and research purposes. Segmenting different parts of the heart provides important diagnostic information. Performing this task manually is time-consuming and operator dependent, hence it would be beneficial to automate this process.

The aim of this thesis is to implement an algorithm to segment the right ventricle in cardiac magnetic resonance images. The right ventricle is geometrically complex, which makes it complicated to introduce prior knowledge to aid the segmentation. Therefore, the chosen segmentation approach was the Active Shape Model (ASM). Based on training data, ASM creates a shape model that will adapt to features in an image,... (More)
Cardiac magnetic resonance is widely used to non-invasive image the heart, for both clinical and research purposes. Segmenting different parts of the heart provides important diagnostic information. Performing this task manually is time-consuming and operator dependent, hence it would be beneficial to automate this process.

The aim of this thesis is to implement an algorithm to segment the right ventricle in cardiac magnetic resonance images. The right ventricle is geometrically complex, which makes it complicated to introduce prior knowledge to aid the segmentation. Therefore, the chosen segmentation approach was the Active Shape Model (ASM). Based on training data, ASM creates a shape model that will adapt to features in an image, making it likely to segment the object of interest.

The project was conducted at the company Medviso AB, and all implementations were done in the software, Segment, a platform for cardiovascular image analysis. The data used included image sets from 115 subjects; 53 heart patients and 62 healthy volunteers. 20% of these image sets were randomly selected as test data, while the rest were used to train the shape model. As validation, segmented volumes were compared to the volumes from manual delineation, and the Dice-Sörensen coefficients were computed. The results were generally good for the mid-ventricular slices, with some errors in the most apical slices.

In conclusion, the presented approach is promising for right ventricle segmentation, but will need further improvements before final implementation
in Segment and clinical usage. (Less)
Please use this url to cite or link to this publication:
author
Gleisner, Gabriella LU
supervisor
organization
course
FMA820 20162
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Active Shape Model, Right Ventricle, Segmentation, Magnetic Resonance Imaging
publication/series
Master's Theses in Mathematical Sciences
report number
LUTFMA-3307-2016
ISSN
1404-6342
other publication id
2016:E51
language
English
id
8905726
date added to LUP
2017-05-29 16:24:32
date last changed
2017-05-29 16:24:32
@misc{8905726,
  abstract     = {Cardiac magnetic resonance is widely used to non-invasive image the heart, for both clinical and research purposes. Segmenting different parts of the heart provides important diagnostic information. Performing this task manually is time-consuming and operator dependent, hence it would be beneficial to automate this process.

The aim of this thesis is to implement an algorithm to segment the right ventricle in cardiac magnetic resonance images. The right ventricle is geometrically complex, which makes it complicated to introduce prior knowledge to aid the segmentation. Therefore, the chosen segmentation approach was the Active Shape Model (ASM). Based on training data, ASM creates a shape model that will adapt to features in an image, making it likely to segment the object of interest.

The project was conducted at the company Medviso AB, and all implementations were done in the software, Segment, a platform for cardiovascular image analysis. The data used included image sets from 115 subjects; 53 heart patients and 62 healthy volunteers. 20% of these image sets were randomly selected as test data, while the rest were used to train the shape model. As validation, segmented volumes were compared to the volumes from manual delineation, and the Dice-Sörensen coefficients were computed. The results were generally good for the mid-ventricular slices, with some errors in the most apical slices.

In conclusion, the presented approach is promising for right ventricle segmentation, but will need further improvements before final implementation
in Segment and clinical usage.},
  author       = {Gleisner, Gabriella},
  issn         = {1404-6342},
  keyword      = {Active Shape Model,Right Ventricle,Segmentation,Magnetic Resonance Imaging},
  language     = {eng},
  note         = {Student Paper},
  series       = {Master's Theses in Mathematical Sciences},
  title        = {Semi-Automatic Segmentation of the Right Ventricle in MR Images},
  year         = {2016},
}