Automatic Left Ventricle Segmentation of Long-Axis Cardiac MRI-Images
(2021) In Master's Theses in Mathematical Sciences FMAM05 20211Mathematics (Faculty of Engineering)
- Abstract
- Over 30 % of all deaths worldwide are related to cardiovascular diseases and they are the leading cause of death. Therefore, research into treatment, diagnosis and prognostication of these diseases are of critical importance.
One way to examine the cardiovascular system is through the use of Magnetic Resonance Imaging (MRI).
The aim of this thesis is to create a model using the Deep Learning architecture known as the U-net to automatically segment the myocardium of the left ventricle of long axis cardiac MRI images. Using this model, the segmentations will be used to calculate the global longitudinal strain (GLS) of the left ventricle.
Several models were created and tested. In total, 601 subjects of varied age and sex were used... (More) - Over 30 % of all deaths worldwide are related to cardiovascular diseases and they are the leading cause of death. Therefore, research into treatment, diagnosis and prognostication of these diseases are of critical importance.
One way to examine the cardiovascular system is through the use of Magnetic Resonance Imaging (MRI).
The aim of this thesis is to create a model using the Deep Learning architecture known as the U-net to automatically segment the myocardium of the left ventricle of long axis cardiac MRI images. Using this model, the segmentations will be used to calculate the global longitudinal strain (GLS) of the left ventricle.
Several models were created and tested. In total, 601 subjects of varied age and sex were used to create and evaluate the models. The data was collected at several different sites using different vendors and settings.
The models were evaluated, and a final model was trained using the optimal settings and parameters. Using the final model, the strain was calculated for all images in the test set and the calculated strain was compared to the strain calculated using the manual delineations.
In conclusion, the final model was able to correctly segment the myocardium of the left ventricle in all views and time steps if given cropped images. The cropping can be done automatically or manually. The segmentations were able to be used for calculations of the global longitudinal strain with a coefficient of variation of about 4.6 % when compared to the manual delineations. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9060726
- author
- Bergengrip, Ruben LU
- supervisor
- organization
- course
- FMAM05 20211
- year
- 2021
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- MRI, Magnetic resonance imaging, Deep Learning, Machine learning, U-net, CNN, Semantic segmentation, Segmentation, Global longitudinal strain, GLS, strain, AV-plane displacement, AVPD, long-axis, left ventricle, LV, cardiac MRI, CMR, Image analysis
- publication/series
- Master's Theses in Mathematical Sciences
- report number
- LUFTMA-3457-2021
- ISSN
- 1404-6342
- other publication id
- 2021:E55
- language
- English
- id
- 9060726
- date added to LUP
- 2021-09-24 16:03:13
- date last changed
- 2021-09-24 16:03:13
@misc{9060726, abstract = {{Over 30 % of all deaths worldwide are related to cardiovascular diseases and they are the leading cause of death. Therefore, research into treatment, diagnosis and prognostication of these diseases are of critical importance. One way to examine the cardiovascular system is through the use of Magnetic Resonance Imaging (MRI). The aim of this thesis is to create a model using the Deep Learning architecture known as the U-net to automatically segment the myocardium of the left ventricle of long axis cardiac MRI images. Using this model, the segmentations will be used to calculate the global longitudinal strain (GLS) of the left ventricle. Several models were created and tested. In total, 601 subjects of varied age and sex were used to create and evaluate the models. The data was collected at several different sites using different vendors and settings. The models were evaluated, and a final model was trained using the optimal settings and parameters. Using the final model, the strain was calculated for all images in the test set and the calculated strain was compared to the strain calculated using the manual delineations. In conclusion, the final model was able to correctly segment the myocardium of the left ventricle in all views and time steps if given cropped images. The cropping can be done automatically or manually. The segmentations were able to be used for calculations of the global longitudinal strain with a coefficient of variation of about 4.6 % when compared to the manual delineations.}}, author = {{Bergengrip, Ruben}}, issn = {{1404-6342}}, language = {{eng}}, note = {{Student Paper}}, series = {{Master's Theses in Mathematical Sciences}}, title = {{Automatic Left Ventricle Segmentation of Long-Axis Cardiac MRI-Images}}, year = {{2021}}, }