A Semi-Automated Approach to Medical Image Segmentation using Conditional Random Field Inference

0

Add to Wishlist
Add to Wishlist

Description

A Semi-Automated Approach to Medical Image Segmentation using Conditional Random Field Inference, Is A Well-Researched Topic, It Is To Be Used As A Guide Or Framework For Your Research

Abstract

Medical image segmentation plays a crucial role in delivering e ective patient care in various diagnostic and treatment modalities. Manual delineation of target volumes and all critical structures is a very tedious and highly time-consuming process and introduce uncertainties
of treatment outcomes of patients. Fully automatic methods holds great promise for reducing cost and time, while at the same time improving accuracy and eliminating expert variability, yet there are still great challenges. Legally and ethically, human oversight must be integrated with “smart tools” favoring a semi-automatic technique which can leverage the best aspects of both human and computer.

In this work we show that we can formulate a semi-automatic framework for the segmentation problem by formulating it as an energy minimization problem in Conditional Random Field (CRF). We show that human input can be used as adaptive training data to condition
a probabilistic boundary term modeled for the heterogeneous boundary characteristics of anatomical structures. We demonstrated that our method can e ortlessly adapt to multiple structures and image modalities using a single CRF framework and tools to learn probabilistic
terms interactively.

To tackle a more dicult multi-class segmentation problem, we
developed a new ensemble one-vs-rest graph cut algorithm. Each graph in the ensemble performs a simple and ecient bi-class (a target class vs the rest of the classes) segmentation.

The nal segmentation is obtained by majority vote. Our algorithm is both faster and more accurate when compared with the prior multi-class method which iteratively swaps classes. In this Thesis, we also include novel volumetric segmentation algorithms which employ deep
learning and indicate how to synthesize our CRF framework with convolutional neural networks (CNN). This would allow incorporating user guidance into CNN based deep learning for this task. We think a deep learning based method interactively guided by human expert
is the ideal solution for medical image segmentation.

Table of Contents

Contents ix
List of Tables xiv
List of Figures xv
1 Introduction 1
2 Medical Image Segmentation Methods 4
2.1 Introduction to medical images segmentation . . . . . . . . . . . . . . . . . . 4
2.2 Problem domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2.1 Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Region based methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4.1 Thresholding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4.2 Region growing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4.3 Region splitting and merging . . . . . . . . . . . . . . . . . . . . . . 13
2.4.4 Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4.5 Bayesian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5 Boundary based methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5.1 Parametric deformable model (active contour) . . . . . . . . . . . . . 20
2.5.2 Non-parametric deformable model (level set, geometric active contour) 22
2.6 Hybrid methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6.1 Level set methods with regional forces . . . . . . . . . . . . . . . . . 25
2.7 Atlas based methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.7.1 Atlas as average shape: active shape model/active appearance model 27
2.7.2 Atlas as individual image with segmentation . . . . . . . . . . . . . . 28
2.7.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.8 Rationale for study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3 Conditional Random Fields and Graph Cut Minimization 36
3.1 Undirected graphical model . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.1.1 Notation and de nitions . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.1.2 Image segmentation with graphical models . . . . . . . . . . . . . . . 38
3.2 Conditional random eld formulation . . . . . . . . . . . . . . . . . . . . . . 39
3.3 s-t cut minimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4 Semi-automatic Bi-class Medical Image Segmentation 45
4.1 Single modality CT liver and kidney segmentation . . . . . . . . . . . . . . . 46
4.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.1.2 Interactive segmentation . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.1.3 Probability estimation . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.1.4 Graph-based contour interpolation . . . . . . . . . . . . . . . . . . . 49
4.1.5 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.1.6 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.1.7 Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.1.8 Acceptability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

4.1.9 Inter- and intra-observer variation . . . . . . . . . . . . . . . . . . . . 54
4.1.10 Time saving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.1.11 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.1.12 Discussion and Conclusions . . . . . . . . . . . . . . . . . . . . . . . 64
4.2 Multi-modality MRI brain tumor segmentation . . . . . . . . . . . . . . . . 67
4.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.2.2 Image features as additional modalities . . . . . . . . . . . . . . . . . 69
4.2.3 Feature selection with random forest . . . . . . . . . . . . . . . . . . 70
4.2.4 Probability estimation . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.2.5 Experiments and results . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.2.6 Conclusions and future works . . . . . . . . . . . . . . . . . . . . . . 75
5 Multi-class Medical Image Segmentation with one-vs-rest graph cuts 81
5.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.2 Multi-class graph cut minimization . . . . . . . . . . . . . . . . . . . . . . . 84
5.2.1 - Swap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.2.2 One-vs-Rest s-t cuts . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.2.3 Majority votes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.3 Experiments and results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.3.1 Data set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.3.2 Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.3.4 Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6 Deep Learning Based Medical Image Segmentation 97

6.1 Multiple Resolution residually connected feature streams for automatic lung
tumor segmentation from CT images . . . . . . . . . . . . . . . . . . . . . . 98
6.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.1.2 Background and motivation . . . . . . . . . . . . . . . . . . . . . . . 100
6.1.3 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.1.4 Implementation details . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.1.5 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.1.6 Evaluation metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.1.7 Preliminary tests to study the in uence of features on segmentation
performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.1.8 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.1.9 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
6.1.10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.2 Tumor-aware, adversarial domain adaptation from CT to MRI for lung cancer
segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.2.2 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.2.3 Step 1: MRI synthesis using tumor-aware unsupervised cross domain
adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
6.2.4 Step 2: Semi-supervised tumor segmentation from MRI . . . . . . . . 131
6.2.5 Network structure and implementation . . . . . . . . . . . . . . . . . 131
6.2.6 Experiments and results . . . . . . . . . . . . . . . . . . . . . . . . . 131
6.2.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.2.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7 Conclusions and Discussions 136

7.1 Summary of thesis contributions . . . . . . . . . . . . . . . . . . . . . . . . . 136
7.1.1 Condition random eld Framework . . . . . . . . . . . . . . . . . . . 137
7.1.2 Multi-class segmentation via ensemble of one-vs-rest graph cuts . . . 139
7.2 Direction for future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Bibliography 143

Brand

YourPastQuestions Brand

Additional information

Author

Yu-chi Hu

No of Chapters

7

No of Pages

177

Reference

YES

Format

PDF

Reviews

There are no reviews yet.

Only logged in customers who have purchased this product may leave a review.