Improvements to Quantification Algorithms for Myocardial Infarction in CMR Images - Validation in Human and Animal Studies
(2013) In Master’s Theses in Mathematical Sciences FMN820 20132Mathematics (Faculty of Engineering)
- Abstract
- Cardiac magnetic resonance (CMR) images are used to investigate the heart for medical and research purposes. By injecting a contrast substance into the patient, myocardial infarctions (heart attacks) can be visualized in CMR image sets consisting of a number of image slices at different levels of the heart. Analysis of these images can detect an infarction, delineate it and estimate its size. This information is then processed by physicians in order to make a diagnosis and decide the course of treatment. Manual delineations are time consuming and observer dependent, why an automated algorithm is desired. Previous work presents a validated automatic segmentation algorithm that calculates a threshold used to separate the healthy tissue... (More)
- Cardiac magnetic resonance (CMR) images are used to investigate the heart for medical and research purposes. By injecting a contrast substance into the patient, myocardial infarctions (heart attacks) can be visualized in CMR image sets consisting of a number of image slices at different levels of the heart. Analysis of these images can detect an infarction, delineate it and estimate its size. This information is then processed by physicians in order to make a diagnosis and decide the course of treatment. Manual delineations are time consuming and observer dependent, why an automated algorithm is desired. Previous work presents a validated automatic segmentation algorithm that calculates a threshold used to separate the healthy tissue pixels from the infarction pixels, based on a fixed number of standard deviations. Theoretically, it is known that algorithms based on standard deviations are likely to be influenced by noise. Therefore, the aim of this thesis was to investigate if other techniques could be used to compute a threshold that is less noise sensitive in both humans and animals.
The study included 40 humans and 18 pigs. Two different techniques based on an Expectation-Maximization algorithm for threshold calculation was developed and implemented into the previous presented method. One implementation analyses each image slice separately (the slice method), and one takes all slices into account at once (the set method). The algorithms were evaluated by comparing computed infarction volume to volumes computed from manual delineations. Both algorithms show good agreement and low bias with the reference standard. The slice method yielded the best results on animal data with a high resolution. The set method yielded the best results in human CMR images, and it show an improved robustness for increasing noise levels. Both implementations show potential for fully automatic quantification of myocardial infarction. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/4128314
- author
- Seemann, Felicia LU
- supervisor
-
- Einar Heiberg LU
- Jane Tufvesson LU
- organization
- alternative title
- Förbättring av kvantifieringsalgoritmer för hjärtinfarkter i MR-bilder - Validering i människor och djur
- course
- FMN820 20132
- year
- 2013
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- EM-algorithm, Myocardial infarction, Segmentation algorithm
- publication/series
- Master’s Theses in Mathematical Sciences
- report number
- LUTFMA-3025-2013
- ISSN
- 1404-6342
- other publication id
- 2013:E58
- language
- English
- id
- 4128314
- date added to LUP
- 2014-02-14 16:19:36
- date last changed
- 2015-12-14 13:32:11
@misc{4128314,
abstract = {{Cardiac magnetic resonance (CMR) images are used to investigate the heart for medical and research purposes. By injecting a contrast substance into the patient, myocardial infarctions (heart attacks) can be visualized in CMR image sets consisting of a number of image slices at different levels of the heart. Analysis of these images can detect an infarction, delineate it and estimate its size. This information is then processed by physicians in order to make a diagnosis and decide the course of treatment. Manual delineations are time consuming and observer dependent, why an automated algorithm is desired. Previous work presents a validated automatic segmentation algorithm that calculates a threshold used to separate the healthy tissue pixels from the infarction pixels, based on a fixed number of standard deviations. Theoretically, it is known that algorithms based on standard deviations are likely to be influenced by noise. Therefore, the aim of this thesis was to investigate if other techniques could be used to compute a threshold that is less noise sensitive in both humans and animals.
The study included 40 humans and 18 pigs. Two different techniques based on an Expectation-Maximization algorithm for threshold calculation was developed and implemented into the previous presented method. One implementation analyses each image slice separately (the slice method), and one takes all slices into account at once (the set method). The algorithms were evaluated by comparing computed infarction volume to volumes computed from manual delineations. Both algorithms show good agreement and low bias with the reference standard. The slice method yielded the best results on animal data with a high resolution. The set method yielded the best results in human CMR images, and it show an improved robustness for increasing noise levels. Both implementations show potential for fully automatic quantification of myocardial infarction.}},
author = {{Seemann, Felicia}},
issn = {{1404-6342}},
language = {{eng}},
note = {{Student Paper}},
series = {{Master’s Theses in Mathematical Sciences}},
title = {{Improvements to Quantification Algorithms for Myocardial Infarction in CMR Images - Validation in Human and Animal Studies}},
year = {{2013}},
}