Artificial Intelligence in Medical Imaging (eBook)

Dr Erik R. Ranschaert, MD, PhD, is currently radiologist at the ETZ Hospital in Tilburg, the Netherlands, and vice-president of the European Society of Medical Imaging Informatics (EuSoMII). Dr. Ranschaert was trained in radiology at  KU Leuven University Hospital in Belgium and graduated in 1994. On July 14th 2016 he was awarded a PhD in Medical Sciences at the University of Antwerp, with a thesis titled: “The Impact of Information Technology on Radiology Services”. He is certified as Imaging Informatics Professional by the ABII in 2017. He was chairman of the ECR Computer Applications Subcommittee in 2008 and member of the ESR eHealth and informatics subcommittee in 2014 - 2016. He is the first author or co-author of more than 20 peer-reviewed articles and he gave more than 40 lectures on invitation, most topics related to his thesis and imaging informatics. Sergey Morozov, MD, MPH, PhD is Professor of Radiology and CEO of Radiology Research and Practice Center in Moscow, Russia. Dr. Morozov was trained in clinical imaging at Sechenov Moscow Medical University and clinical effectiveness at Harvard School of Public Health in 2002-2006. He became Chief of Radiology at the Central Clinical Hospital in Moscow in 2007-2012 and then at the European Medical Center in 2013-2015. He is Executive Director of Russian Society of Radiology, President of European Society of Medical Imaging Informatics, past chairman of Imaging Informatics subcommittee of ECR, member of ECR 2019 Program planning committee, RSNA Education Exhibits Awards Committee. He is certified as Imaging Informatics Professional by ABII in 2017. Prof. Dr. Morozov is a renowned expert in clinical imaging, healthcare management and informatics and is the co-author of more than 100 journal articles and 15 book chapters. Dr Paul Algra MD PhD, Northwest Hospital Group, Alkmaar, The Netherlands, is trained as radiologist in Leiden University Hospital and as neuroradiologist in Free University Amsterdam. His PhD thesis (1992) was on CT and MRI of vertebral metastases. He was vice-president of Dutch Radiological Society. He is member of scientific committee CAR, board member of EuSoMII and editorial board member of several radiology journals. He (co) authored around 50 articles in peer reviewed journals and served as department chief and program director for more than 15 years.

I've Seen the Future … 5
Preface 9
Contents 12
Part I Introduction 15
1 Introduction: Game Changers in Radiology 16
1.1 Era of Changes 16
1.2 Perspectives 17
1.3 Opportunities for the Future 17
1.4 Conclusion 17
Reference 18
Part II Technology: Getting Started 19
2 The Role of Medical Image Computing and Machine Learning in Healthcare 20
2.1 Introduction 20
2.2 Medical Image Analysis 20
2.2.1 Image Segmentation 21
2.2.2 Image Registration 21
2.2.3 Image Visualization 22
2.3 Challenges 23
2.3.1 Complexity of the Data 23
2.3.2 Complexity of the Objects of Interest 23
2.3.3 Complexity of the Validation 23
2.4 Medical Image Computing 24
2.5 Model-Based Image Analysis 25
2.5.1 Energy Minimization 25
2.5.2 Classification/Regression 26
2.6 Computational Strategies 27
2.6.1 Flexible Shape Fitting 28
2.6.2 Pixel Classification 30
2.7 Fundamental Issues 31
2.7.1 Explicit Versus Implicit Representation of Geometry 32
2.7.2 Global Versus Local Representations of Appearance 32
2.7.3 Deterministic Versus Statistical Models 33
2.7.4 Data Congruency Versus Model Fidelity 33
2.8 Conclusion 34
References 34
3 A Deeper Understanding of Deep Learning 35
3.1 Introduction 35
3.2 Computer-Aided Diagnosis, the Classical Approaches 36
3.3 Artificial Intelligence 36
3.4 Neural Networks 36
3.5 Convolutional Neural Networks 38
3.6 Why Now? 40
3.7 Example: Screening for Diabetic Retinopathy 40
3.8 Pointers on the Web 41
3.9 A Comparison with Brain Research 42
3.9.1 Brain Efficiency 42
3.9.2 Visual Learning 42
3.9.3 Foveated Vision 44
3.10 Conclusions and Recommendations 45
3.11 Take Home Messages 47
References 47
4 Deep Learning and Machine Learning in Imaging: Basic Principles 49
4.1 Introduction 49
4.2 Features and Classes 49
4.3 Neural Networks 50
4.4 Support Vector Machines 51
4.5 Decision Trees 51
4.6 Bayes Network 52
4.7 Deep Learning 52
4.7.1 Deep Learning Layers 53
4.7.2 Deep Learning Architectures 54
4.8 Conclusion 55
References 55
Part III Technology: Developing A.I. Applications 57
5 How to Develop Artificial Intelligence Applications 58
5.1 Introduction 58
5.2 Applications of AI in Radiology 59
5.3 Development of AI Applications in Radiology 63
5.4 Resources Framework 65
5.5 Conclusion 67
5.6 Summary/Take-Home Points 67
References 68
6 A Standardised Approach for Preparing Imaging Data for Machine Learning Tasks in Radiology 69
6.1 Data, Data Everywhere? 69
6.2 Not All Data Is Created Equal 70
6.3 The MIDaR Scale 71
6.3.1 MIDaR Level D 72
6.3.2 MIDaR Level C 74
6.3.3 MIDaR Level B 75
6.3.4 MIDaR Level A 77
6.4 Summary 78
6.5 Take Home Points 79
References 79
7 The Value of Structured Reporting for AI 81
7.1 Introduction 81
7.2 Conventional Radiological Reporting Versus Structured Reporting 82
7.3 Technical Implementations of Structured Reporting and IHE MRRT 83
7.4 Information Extraction Using Natural Language Processing 84
7.5 Information Extraction from Structured Reports 85
7.6 Integration of External Data into Structured Reports 86
7.7 Analytics and Clinical Decision Support 86
7.8 Outlook 88
References 88
8 Artificial Intelligence in Medicine: Validation and Study Design 91
8.1 The Validation of AI Technologies in Medicine 91
8.2 Safety in Medical AI 92
8.3 Assessing Model Efficacy Using Clinical Studies 93
8.3.1 The Clinical Question 95
8.3.2 The Ground Truth 95
8.3.3 The Target Population 97
8.3.4 The Cohort 98
8.3.5 Metrics 101
8.3.6 The Analysis 103
8.4 An Example of Study Design 107
8.5 Assessing Safety in Medical AI 108
8.6 Take-Home Points 110
References 111
Part IV Big Data in Medicine 113
9 Enterprise Imaging 114
9.1 Introduction 114
9.2 Basic Principles of Enterprise Imaging (EI) 115
9.3 Enterprise Imaging Platform 116
9.4 Standards and Technology for an Enterprise Imaging Platform and Image Sharing Across Enterprises 119
9.5 Legal Aspects 120
9.6 Enterprise Imaging in the Context of Artificial Intelligence 121
9.7 Take-Home Points 122
References 123
10 Imaging Biomarkers and Imaging Biobanks 125
10.1 Introduction 125
10.2 Stepwise Development 126
10.3 Validation 127
10.4 Imaging Biobanks 129
10.5 Conclusion 131
10.6 Take-Home Points 131
References 131
Part V Practical Use Cases of A.I. in Radiology 133
11 Applications of AI Beyond Image Interpretation 134
11.1 Imaging Appropriateness and Utilization 135
11.2 Patient Scheduling 135
11.3 Imaging Protocoling 136
11.4 Image Quality Improvement and Acquisition Time Reduction in MRI 136
11.5 Image Quality Improvement and Radiation Dose Reduction 137
11.6 Image Transformation 137
11.7 Image Quality Evaluation 137
11.8 Hanging Protocols 138
11.9 Reporting 138
11.10 Text Summarization and Report Translation 139
11.11 Speech Recognition 140
11.12 Follow-up 140
11.13 Worklist Optimization 140
11.14 Staffing Optimization 141
11.15 Business Intelligence and Business Analytics 141
11.16 Content-Based Image Retrieval 141
11.17 Patient Safety 142
11.18 Billing 142
11.19 Patient Experience 142
11.20 Challenges 143
11.21 Conclusion 143
11.22 Take-Home Points 143
References 144
12 Artificial Intelligence and Computer-Assisted Evaluation of Chest Pathology 149
12.1 Introduction 149
12.2 General Chest Radiography 149
12.3 Lung Nodules 150
12.3.1 Chest Radiography 151
12.3.2 Computed Tomography 152
12.4 Lung Cancer Radiomics 157
12.5 Pulmonary Embolism 159
12.6 Parenchymal Lung and Airways Diseases 161
12.7 Interstitial Lung Disease 163
12.8 Conclusions 165
12.9 Take-Home Points 167
References 167
13 Cardiovascular Diseases 171
13.1 Introduction 171
13.2 Impact of AI on Cardiovascular Imaging 172
13.2.1 Decision Support 172
13.2.2 Image Acquisition 173
13.2.3 Image Reconstruction and Improvement of Image Quality 173
13.2.4 Post-processing and Image Analysis 173
13.2.5 Interpretation and Diagnosis 173
13.2.6 Opportunistic Screening and Prognosis 174
13.2.7 Combining Imaging with Other Data Sources 174
13.3 Practical Use of AI in Different Cardiovascular Imaging Modalities 174
13.3.1 Echocardiography 174
13.3.2 Computed Tomography 176
13.3.3 Magnetic Resonance Imaging 179
13.3.4 Nuclear Imaging 181
13.3.5 Outcome Prediction Based on Composite Data 182
13.3.6 Deployment of Algorithms in Clinical Practice 183
13.3.7 Outlook and Conclusions 184
References 185
14 Deep Learning in Breast Cancer Screening 190
14.1 Background 190
14.1.1 The Breast Cancer Screening Global Landscape 190
14.1.2 The Rise and Fall of CAD 192
14.1.2.1 Rise: The Premise and Promise 192
14.1.2.2 How Does CAD Perform? 192
14.1.2.3 So Why Did CAD ``Fail''? 195
14.1.3 A Brief History of Deep Learning for Mammography 196
14.2 Goals for Automated Systems 197
14.2.1 Recall Decision Support 197
14.2.2 Lesion Localization 198
14.2.3 Density Stratification and Risk Prediction 200
14.3 Deep Learning Challenges Specific to Mammography 202
14.3.1 Memory Constraints and Image Size 202
14.3.2 Data Access and Quality 203
14.3.3 Data Issues During Training 204
14.3.3.1 Dataset Imbalance 204
14.3.3.2 Dataset Bias 205
14.3.3.3 Under-Fitting, Overfitting, and Generalization 205
14.3.4 Data Labeling 206
14.3.5 Principled Uncertainties 207
14.3.6 Interpretability 207
14.4 Future Directions 209
14.4.1 Generative Adversarial Networks (GANs) 209
14.4.2 Active Learning and Regulation 210
14.4.3 Tomosynthesis 211
14.4.4 Genomics 212
14.5 Summary 212
14.6 Take Home Points 213
References 213
15 Neurological Diseases 219
15.1 Introduction 219
15.2 Preprocessing of Brain Imaging 219
15.3 Applications 220
15.3.1 Protocoling, Acquisition, and Image Construction 220
15.3.2 Segmentation 222
15.3.3 Stroke 223
15.3.4 Tumor Classification 224
15.3.5 Disease Detection 225
15.4 Conclusion 226
15.5 Take-Home Points 227
References 227
16 The Role of AI in Clinical Trials 233
16.1 Introduction 233
16.2 Standardization of Medical Imaging in Clinical Trials 234
16.2.1 Before the Start of Clinical Trial 235
16.2.1.1 Image Acquisition Protocol Design 235
16.2.1.2 Site Validation 235
16.2.1.3 During the Clinical Trial 237
16.3 Artificial Intelligence in Clinical Trials 238
16.3.1 Classification Algorithms 241
16.3.1.1 Segmentation Algorithms 241
16.4 Digital Twin and In Silico Clinical Trials 242
16.5 Conclusion 244
16.6 Summary 244
References 244
Part VI Quality, Regulatory and Ethical Issues 246
17 Quality and Curation of Medical Images and Data 247
17.1 Introduction 247
17.2 Data Discovery and Retrieval 249
17.3 Data Quality 251
17.4 Adding Value 252
17.5 Reuse Over Time 253
17.6 Some Tools of the Trade 253
17.7 Conclusions 253
References 254
18 Does Future Society Need Legal Personhood for Robots and AI? 256
18.1 A Paradigm Shift 256
18.2 Legal Position 259
18.3 AI and Robots as Actor 260
18.4 Legal Subjectivity 261
18.5 Humans as (Natural) Legal Persons 263
18.5.1 Human-Like Behavior as Determination for Legal Personhood 266
18.5.2 Non-natural (Artificial) Legal Persons 268
18.6 Autonomous Artificial Intelligent Entities 269
18.6.1 AI in Robotic Entities 269
18.7 The Problem of Human–Robot Integration 272
18.8 An Alternative Personhood 274
18.8.1 Abstraction of the Legal Position of the Robot by a Narrative 275
18.8.1.1 The Cheshire Cat 275
18.8.1.2 The Reasonable Human 275
18.8.1.3 The Responsible Actor 276
18.8.1.4 Concluding on Legal Position 277
18.9 The Artificial Intelligent Entity or Robot as Legal Actor 279
18.9.1 Sui Generis Construct, Legal Subject or Legal Object Specialis? 280
18.9.2 Liability and Legal Subjectivity 280
18.9.3 Legal Acts 282
18.10 Where to Go from Here? 284
References 288
19 The Role of an Artificial Intelligence Ecosystem in Radiology 290
19.1 Defining Business Ecosystems 290
19.2 Artificial Intelligence Ecosystem for Healthcare and Diagnostic Imaging 292
19.3 Defining an Artificial Intelligence Ecosystem in Healthcare with a Focus on Diagnostic Imaging 294
19.3.1 Establish Realistic Goals 294
19.3.2 Maintain a Targeted Focus 296
19.3.3 Use High-Quality Datafor Training and Testing 298
19.3.4 Develop Consistent Methods for Validation and Monitoring Algorithm Performance 299
19.3.5 Build Public-Private Partnerships for Safety and Efficacy 300
19.3.6 Establish Standardsfor Interoperability and Pathways for Integration into Clinical Workflows 302
19.3.7 Promote Explicability of Algorithm Output 306
19.3.8 Facilitate Radiologist Input into Development, Validation, and Implementation 307
19.4 Bringing Artificial Products to Widespread Clinical Use: Challenges, Opportunities for Radiologists, and the Role of Medical Specialty Societies 308
19.4.1 Creating Clinically Effective Artificial Intelligence Use Cases 309
19.4.2 Enhancing the Availability of High-Quality Datasetsfor Algorithm Testing and Training 312
19.4.3 Maintaining Patient Data Privacy in Developing and Validating Artificial Intelligence Algorithms 314
19.4.4 Enhancing Algorithm Validation 315
19.4.5 Enhancing Clinical Integration 316
19.4.6 Mechanisms for Assessing Algorithm Performance in Clinical Practice 316
19.4.7 The Economics of AI and Business Models for Moving AI to Clinical Practice 317
19.4.8 Facilitating the Development of Non-interpretive Use Cases for Artificial Intelligence in Radiological Practice 319
19.4.9 Educating Non-radiologist Stakeholders About the Value of AI 319
19.5 Summary of the ProposedAI Ecosystemfor the Radiological Sciences 320
19.6 Conclusion 322
References 323
20 Advantages, Challenges, and Risks of Artificial Intelligencefor Radiologists 327
20.1 Innovation in Radiology 327
20.1.1 Artificial Intelligence (AI) Is the Next Big Thing 328
20.1.2 Radiologists' Perspective 329
20.2 Level of Expectation for AI in Radiology 331
20.2.1 AI Will Complement Many Routine Radiology Tasks 331
20.2.2 Will AI Also Surpass Existing Radiology Tasks? 333
20.3 Strategies to Prepare for the Future 335
20.3.1 Multitask Learning 335
20.3.2 Swiss Knife for Radiologists 336
20.3.3 Integration of Existing Medical Information Databases 336
20.3.4 Blockchain Technology 337
20.4 Hidden Risks and Dangers 338
20.4.1 Quality and Validation of Data 339
20.4.2 Data Security and Privacy 339
20.4.3 Ethics and AI 340
20.5 Take-Home Messages 342
References 343
AI: A Glossary of Terms 345
Glossary 346
Index 361