Deep Learning and Convolutional Neural Networks for Medical Image Computing (eBook)

Dr. Le Lu is a Staff Scientist in the Radiology and Imaging Sciences Department of the National Institutes of Health Clinical Center, Bethesda, MD, USA.Dr. Yefeng Zheng is a Senior Staff Scientist at Siemens Healthcare Technology Center, Princeton, NJ, USA.Dr. Gustavo Carneiro is an Associate Professor in the School of Computer Science at The University of Adelaide, Australia.Dr. Lin Yang is an Associate Professor in the Department of Biomedical Engineering at the University of Florida, Gainesville, FL, USA.

Preface 6
Overview and Goals 7
Organization and Features 8
Target Audience 9
Acknowledgements 10
Contents 11
Part I Review 14
1 Deep Learning and Computer-Aided Diagnosis for Medical Image Processing: A Personal Perspective 15
References 20
2 Review of Deep Learning Methods in Mammography, Cardiovascular, and Microscopy Image Analysis 23
2.1 Introduction on Deep Learning Methods in Mammography 23
2.2 Deep Learning Methods in Mammography 24
2.3 Summary on Deep Learning Methods in Mammography 26
2.4 Introduction on Deep Learning for Cardiological Image Analysis 26
2.5 Deep Learning-Based Methods for Heart Segmentation 28
2.6 Deep Learning-Based Methods for Vessel Segmentation 29
2.7 Introduction to Microscopy Image Analysis 31
2.8 Deep Learning Methods 33
2.9 Microscopy Image Analysis Applications 34
2.10 Discussions and Conclusion on Deep Learning for Microscopy Image Analysis 34
References 38
Part II Detection and Localization 45
3 Efficient False Positive Reduction in Computer-Aided Detection Using Convolutional Neural Networks and Random View Aggregation 46
3.1 Introduction 47
3.2 Related Work 47
3.2.1 Cascaded Classifiers in CADe 48
3.3 Methods 48
3.3.1 Convolutional Neural Networks 48
3.3.2 A 2D or 2.5D Approach for Applying ConvNets to CADe 50
3.3.3 Random View Aggregation 52
3.3.4 Candidate Generation 52
3.4 Results 53
3.4.1 Computer-Aided Detection Data Sets 53
3.5 Discussion and Conclusions 54
References 56
4 Robust Landmark Detection in Volumetric Data with Efficient 3D Deep Learning 60
4.1 Introduction 60
4.2 Training Shallow Network with Separable Filters 63
4.3 Training Sparse Deep Network 66
4.4 Robust Detection by Combining Multiple Features 67
4.5 Experiments 68
4.6 Conclusions 70
References 71
5 A Novel Cell Detection Method Using Deep Convolutional Neural Network and Maximum-Weight Independent Set 73
5.1 Introduction 73
5.2 Methodology 75
5.2.1 Cell Detection Using MWIS 75
5.2.2 Deep Convolutional Neural Network 76
5.3 Experiments 78
5.4 Conclusion 81
References 81
6 Deep Learning for Histopathological Image Analysis: Towards Computerized Diagnosis on Cancers 83
6.1 Introduction 84
6.2 Previous Works 85
6.2.1 Previous Works on Deep Learning for Histological Image Analysis 86
6.2.2 Previous Works on Nuclear Atypia Scoring 87
6.2.3 Previous Works on Epithelial and Stromal Segmentation 88
6.3 Deep Learning for Nuclear Atypia Scoring 88
6.3.1 CN Model for Nuclear Atypia Scoring 90
6.3.2 Integration MR-CN with Combination Voting Strategies for NAS 91
6.4 Deep Learning for Epithelial and Stromal Tissues Segmentation 94
6.4.1 The Deep Convolutional Neural Networks 94
6.4.2 Generating Training and Testing Samples 94
6.4.3 The Trained CN for the Discrimination of EP and ST Regions 95
6.5 Experimental Setup 96
6.5.1 Data Set 97
6.5.2 Comparison Strategies 98
6.5.3 Computational and Implemental Consideration 99
6.6 Results and Discussion 99
6.6.1 Qualitative Results 99
6.6.2 Quantitative Results 100
6.7 Concluding Remarks 102
References 102
7 Interstitial Lung Diseases via Deep Convolutional Neural Networks: Segmentation Label Propagation, Unordered Pooling and Cross-Dataset Learning 106
7.1 Introduction 106
7.2 Methods 109
7.2.1 Segmentation Label Propagation 110
7.2.2 Multi-label ILD Regression 112
7.3 Experiments and Discussion 114
7.3.1 Segmentation Label Propagation 114
7.3.2 Multi-label ILD Regression 116
7.4 Conclusion 118
References 119
8 Three Aspects on Using Convolutional Neural Networks for Computer-Aided Detection in Medical Imaging 121
8.1 Introduction 122
8.2 Datasets and Related Work 124
8.3 Methods 127
8.3.1 Convolutional Neural Network Architectures 127
8.3.2 ImageNet: Large-Scale Annotated Natural Image Dataset 129
8.3.3 Training Protocols and Transfer Learning 129
8.4 Experiments and Discussions 131
8.4.1 Thoracoabdominal Lymph Node Detection 131
8.4.2 Interstitial Lung Disease Classification 133
8.4.3 Evaluation of Five CNN Models Using ILD Classification 137
8.4.4 Analysis via CNN Learning Visualization 138
8.4.5 Findings and Observations 139
8.5 Conclusion 140
References 141
9 Cell Detection with Deep Learning Accelerated by Sparse Kernel 145
9.1 Introduction 145
9.1.1 Related Work 146
9.1.2 Challenges 152
9.2 Pixel-Wise Cell Detector 153
9.2.1 Overview 153
9.2.2 Deep Convolutional Neural Network 154
9.2.3 Implementation 155
9.3 Sparse Kernel Acceleration of the Pixel-Wise Cell Detector 156
9.3.1 Training the Detector 156
9.3.2 Deep Convolution Neural Network Architecture 156
9.3.3 Acceleration of Forward Detection 157
9.4 Experiments 158
9.4.1 Materials and Experiment Setup 158
9.4.2 Results 160
9.5 Discussion 160
9.6 Conclusion 162
References 163
10 Fully Convolutional Networks in Medical Imaging: Applications to Image Enhancement and Recognition 166
10.1 Introduction 166
10.2 Image Super-Resolution 168
10.2.1 Motivation 168
10.2.2 Methodology 169
10.2.3 Results 172
10.2.4 Discussion and Conclusion 175
10.3 Scan Plane Detection 175
10.3.1 Motivation 175
10.3.2 Materials and Methods 177
10.3.3 Experiments and Results 181
10.3.4 Discussion and Conclusion 182
10.4 Discussion and Conclusion 184
References 185
11 On the Necessity of Fine-Tuned Convolutional Neural Networks for Medical Imaging 187
11.1 Introduction 188
11.2 Related Works 188
11.3 Contributions 190
11.4 Applications and Results 190
11.4.1 Polyp Detection 192
11.4.2 Pulmonary Embolism Detection 193
11.4.3 Colonoscopy Frame Classification 194
11.4.4 Intima-Media Boundary Segmentation 195
11.5 Discussion 196
11.6 Conclusion 197
References 198
Part III Segmentation 200
12 Fully Automated Segmentation Using Distance Regularised Level Set and Deep-Structured Learning and Inference 201
12.1 Introduction 202
12.2 Literature Review 203
12.3 Methodology 205
12.3.1 Left Ventricle Segmentation 208
12.3.2 Endocardium Segmentation 209
12.3.3 Epicardium Segmentation 213
12.3.4 Lung Segmentation 214
12.4 Experiments 214
12.4.1 Data Sets and Evaluation Measures 214
12.4.2 Experimental Setup 215
12.4.3 Results of Each Stage of the Proposed Methodology 217
12.4.4 Comparison with the State of the Art 217
12.5 Discussion and Conclusions 219
References 227
13 Combining Deep Learning and Structured Prediction for Segmenting Masses in Mammograms 229
13.1 Introduction 229
13.2 Literature Review 231
13.3 Methodology 232
13.3.1 Conditional Random Field (CRF) 233
13.3.2 Structured Support Vector Machine (SSVM) 235
13.3.3 Potential Functions 236
13.4 Experiments 239
13.4.1 Materials and Methods 239
13.4.2 Results 240
13.5 Discussion and Conclusions 242
References 243
14 Deep Learning Based Automatic Segmentation of Pathological Kidney in CT: Local Versus Global Image Context 245
14.1 Introduction 245
14.2 Training Data Synthesis 248
14.3 Abdomen Localization Using Deep Learning 250
14.4 Kidney Localization Using Deep Learning 253
14.5 Kidney Segmentation Based on MSL 254
14.6 Experiments 255
14.7 Conclusions 258
References 258
15 Robust Cell Detection and Segmentation in Histopathological Images Using Sparse Reconstruction and Stacked Denoising Autoencoders 260
15.1 Introduction 261
15.2 Methodology 263
15.2.1 Detection via Sparse Reconstruction with Trivial Templates 264
15.2.2 Cell Segmentation via Stacked Denoising Autoencoders 265
15.2.3 The Learned Filters 267
15.2.4 Training DAE with Discriminative Loss 268
15.3 Experimental Results 273
15.3.1 Computational Complexity 278
15.4 Conclusion 279
References 279
16 Automatic Pancreas Segmentation Using Coarse-to-Fine Superpixel Labeling 282
16.1 Introduction 283
16.2 Previous Literature 285
16.3 Methods 286
16.3.1 Boundary-Preserving Over-segmentation 286
16.3.2 Patch-Level Visual Feature Extraction and Classification: PRF 289
16.3.3 Patch-Level Labeling via Deep Convolutional Neural Network: PCNN 292
16.3.4 Superpixel-Level Feature Extraction, Cascaded Classification, and Pancreas Segmentation 293
16.4 Data and Experimental Results 295
16.4.1 Imaging Data 295
16.4.2 Experiments 296
16.5 Conclusion and Discussion 302
References 303
Part IV Big Dataset and Text-Image Deep Mining 306
17 Interleaved Text/Image Deep Mining on a Large-Scale Radiology Image Database 307
17.1 Introduction 307
17.1.1 Related Work 309
17.2 Data 310
17.3 Document Topic Learning with Latent Dirichlet Allocation 311
17.4 Image to Document Topic Mapping with Deep Convolutional Neural Networks 313
17.5 Generating Image-to-Text Description 315
17.5.1 Removing Word-Level Ambiguity with Word-to-Vector Modeling 316
17.5.2 Using Sentences to Words Based Image Representation 317
17.5.3 Bi-gram Deep CNN Regression 317
17.5.4 Word Prediction from Images as Retrieval 318
17.6 Conclusion and Discussion 320
References 321
Author Index 324
Subject Index 326