Segmentation of Cardiac Delayed Enhancement Magnetic Resonance Images / Mostafa Ahmed Ibrahim AlAttar

Supervisor: Nael Osman

Thesis (M.A.)—Nile University, Egypt, 2010 .

"Includes bibliographical references"

Contents:
INTRODUCTION .............................................................................................................. 1
1.1. Motivation ...................................................................................................... 1
1.2. Problem Statement ......................................................................................... 4
1.3. Thesis Objective ............................................................................................. 4
1.4. Thesis Overview ............................................................................................. 5
BACKGROUND ................................................................................................................ 5
2.1. Physiology Review ......................................................................................... 5
2.1.1. Heart Anatomy ........................................................................................ 5
2.1.2. Physiological Measurements .................................................................. 7
2.2. MRI concepts and data material ..................................................................... 8
2.2.1. History of MRI ........................................................................................ 8
2.2.3. Cardiac MRI.......................................................................................... 20
2.2.4. Cardiac MRI Imaging Methods: ........................................................... 21
RELATED WORK ........................................................................................................... 26
3.1. Thresholding ................................................................................................. 26
3.2. Region Growing ........................................................................................... 28
3.3. Classifiers ..................................................................................................... 30
3.4. Clustering ..................................................................................................... 33
3.5. Graph-based segmentation ........................................................................... 35
3.6. Active contour models ACMs ...................................................................... 35
3.7. Active shape models ASMs ......................................................................... 36
3.8. Active appearance models AAMs ................................................................ 37
3.9. Level-set segmentation ................................................................................. 37
CINE BRIGHT BLOOD IMAGES SEGMENTATION METHODS.............................. 38
4.1. Multi Seeded Region Growing (Experiment 1) ........................................... 38
4.1.1. Standard region growing algorithm (RG) ............................................. 38
4.1.2. Limitations of region growing .............................................................. 39
4.1.3. Multi-Seeded Region Growing (MSRG) .............................................. 40
xvi
4.1.4. Morphological Operations .................................................................... 44
4.1.5. Epicardial Control Points ...................................................................... 48
4.1.6. Refinement of the Contours using ACMs ............................................. 51
4.1.7. Proposed Algorithm .............................................................................. 54
4.2. Adaptive Multi Seeded Region Growing (Experiment 2) ............................ 55
4.2.1. Multi-Seeded Region Growing ............................................................. 55
4.2.2. Seed Points and Constraining Areas Selection ..................................... 55
4.2.3. K-Means ................................................................................................ 55
4.2.4. Proposed Algorithm .............................................................................. 57
4.3. Automatic Evaluation Methods .................................................................... 58
4.3.1. Average perpendicular distance ............................................................ 58
4.3.2. Dice metric ............................................................................................ 60
4.3.3. Accuracy Measurements ....................................................................... 61
4.3.4. Visual (Manual) Evaluation Method .................................................... 61
4.3.5. Another Parameters that’s taken into account ...................................... 61
CINE BLACK BLOOD IMAGES CLASSIFICATION .................................................. 62
5.1. Modified STEAM for Black Blood Imaging ............................................... 62
5.2. Stochastic Model .......................................................................................... 63
5.3. Classification Techniques ............................................................................ 65
5.3.1. Bayesian Classifier................................................................................ 65
5.3.2. Linear Classifier L1-Norm .................................................................... 65
5.3.3. Quadratic Classifier L2-Norm ............................................................... 66
5.3.4. Rectangular Classifier Linf-Norm .......................................................... 67
5.4. Numerical Simulation .................................................................................. 67
5.4.1. Case1: Background Signal .................................................................... 68
5.4.2. Case2: Tissue Signal ............................................................................. 69
5.5. Real Images Test .......................................................................................... 70
5.6. Further Processing ........................................................................................ 72
5.7. Performance Evaluation Methods ................................................................ 74
CINE BRIGHT BLOOD SEGMENTATION RESULTS & DISCUSSION ................... 75
6.1. Multi Seeded Region Growing (Experiment 1) Results............................... 75
6.1.1. Simple Region Growing ....................................................................... 76
6.1.2. Multi Seeded Region Growing constrained by Overlapped Sectors .... 77
xvii
6.1.3. Evaluation of the MSRG Coefficients .................................................. 77
6.1.4. Resulted Contour Refinement using ACM and CPR ............................ 80
6.1.5. Final Results.......................................................................................... 80
6.2. Adaptive Multi Seeded Region Growing (Experiment 2) Results ............... 85
6.2.1. Selecting the sectors sizes and positions using k-means ...................... 85
6.2.2. Quantitative Analysis ............................................................................ 86
6.2.3. Final Results: ........................................................................................ 87
CINE BLACK BLOOD CLASSIFICATION RESULTS & DISCUSSION ................... 92
7.1. Simulated Data Results ................................................................................ 92
7.2. Real Images Results ..................................................................................... 95
CONCLUSION ............................................................................................................... 101
8.1. Future Work ............................................................................................... 103
REFERENCES ...............................................................................................................

Abstract:
Imaging the heart using Cine Bright Blood MRI and Cine Black-blood MRI
sequences is very important to evaluate the cardiac global and regional function. In
the first type of imaging which is Cine Bright Blood MRI, Manual segmentation of
the contours in all images through different slices is a cumbersome task. Therefore,
methods were proposed to automatically or semi-automatically analyze and segment
the contours from short-axis images and derive useful clinical information from them
is highly desirable. In this thesis, we have proposed two algorithms to segment LV
and they are Multi-Seeded Region Growing (MSRG) and adaptive MSRG. Their
performance has been evaluated also in this thesis. The second type of images which
is Cine Black Blood images suffers from the low signal-to-noise ratio SNR in
general. In this thesis, a probabilistic model of blood and tissue signals is developed
and used to build a Bayesian decision function to identify and filter the noise from the
background signal. Numerical simulation and real MRI images were used to test and
validate the proposed method. Also, the proposed method is compared to other
conventional techniques.

Text in English, abstracts in English.