Introduction to Artificial Intelligence

This book aims to provide physicians and scientists with the basics of Artificial Intelligence (AI) with a special focus on medical imaging. The contents of the book provide an introduction to the main topics of artificial intelligence currently applied on medical image analysis. The book starts with a chapter explaining the basic terms used in artificial intelligence for novice readers and embarks on a series of chapters each one of which provides the basics on one AI-related topic. The second chapter presents the programming languages and available automated tools that enable the development of AI applications for medical imaging. The third chapter endeavours to analyse the main traditional machine learning techniques, explaining algorithms such as random forests, support vector machines as well as basic neural networks. The applications of those machines on the analysis of radiomics data is expanded in the fourth chapter to allow the understanding of algorithms used to build classifiers for the diagnosis of disease processes with the use of radiomics. Chapter five provides the basics of natural language processing which has revolutionized the analysis of complex radiological reports and chapter six affords a succinct introduction to convolutional neural networks which have revolutionized medical image analysis enabling automated image-based diagnosis, image enhancement (e.g. denoising), protocolling etc. The penultimate chapter provides an introduction to data preprocessing for use in the aforementioned artificial intelligence applications. The book concludes with a chapter demonstrating AI-based tools already in radiological practice while providing an insight about the foreseeable future. It will be a valuable resource for radiologists, computer scientists and postgraduate students working on medical image analysis. Preface 6 Contents 8 1 What Is Artificial Intelligence: History and Basic Definitions 10 1.1 Twentieth Century: Setting the Foundations of Artificial Intelligence 10 1.1.1 Artificial Intelligence 10 1.1.2 Machine Learning 12 1.1.2.1 Neural Networks 14 1.2 The Period 2000–2020 16 References 20 2 Using Commercial and Open-Source Tools for Artificial Intelligence: A Case Demonstration on a Complete Radiomics Pipeline 21 2.1 Introduction 22 2.2 Image Segmentation 23 2.3 Image Pre-processing 25 2.4 Radiomics Extraction 27 2.5 Radiomics Modeling 28 2.6 From Theory to Practice 30 2.7 Discussion 35 2.8 Conclusion 36 References 37 3 Introduction to Machine Learning in Medicine 46 3.1 Introduction 46 3.2 What Is Machine Learning? 47 3.3 Principal ML Algorithms 52 3.3.1 Supervised Machine Learning 52 3.3.1.1 Linear Regression 55 3.3.1.2 Support Vector Machine 55 3.3.1.3 Random Decision Forest 56 3.3.1.4 Extreme Gradient Boosting 56 3.3.1.5 Naive Bayes 56 3.3.2 Unsupervised Machine Learning 57 3.3.2.1 k-Nearest Neighbours 59 3.3.2.2 Principal Component Analysis 59 3.3.2.3 k-Means Clustering 59 3.3.3 Artificial Neural Networks 60 3.3.4 Reinforcement Learning 60 3.4 Issues and Challenges 61 3.4.1 Data Management 61 3.4.2 Machine Learning Model Evaluation Metrics 61 3.4.3 Explainability, Interpretability, and Ethical and Legal Issues 63 3.4.4 Perspectives in Personalized Medicine 64 3.5 Conclusions 65 References 67 4 Machine Learning Methods for Radiomics Analysis: Algorithms Made Easy 76 4.1 Introduction 76 4.2 Methods for Region of Interest Segmentation 78 4.2.1 R-CNN 79 4.2.2 U-Net and V-Net 80 4.2.3 DeepLab 80 4.3 Methods for Exploratory Data Analysis 81 4.3.1 Correlation Analysis 81 4.3.2 Clustering 82 4.3.3 Principal Component Analysis 83 4.4 Methods for Feature Selection 83 4.4.1 Boruta 84 4.4.2 Recursive Feature Elimination 85 4.4.3 Maximum Relevance: Minimum Redundancy 86 4.5 Methods for Predictive Model Construction 86 4.5.1 Decision Trees 87 4.5.2 Random Forests 87 4.5.3 Gradient Boosting Algorithms 88 4.5.4 Support Vector Machines 89 4.5.5 Neural Networks 89 4.6 Conclusion 89 References 90 5 Natural Language Processing 93 5.1 Brief History of NLP 93 5.2 Basic of Natural Language Processing 96 5.3 Current Applications of Natural Language Processing 99 References 102 6 Deep Learning Fundamentals 106 Abbreviations 106 6.1 Deep Learning in Medical Imaging 107 6.1.1 Key Concepts 107 6.1.2 DL Architectures for Medical Image Analysis*-9pt 109 6.1.3 Cloud Computing for Deep Learning 113 6.1.4 DL-Based Computer-Aided Diagnosis 114 6.2 Quality and Biases of Medical Databases 115 6.3 Pre-processing for Deep Learning 117 6.3.1 CT Radiation Absorption Map to Grayscale 117 6.3.2 MRI Bias Field Correction 117 6.3.3 Tissue-Based Standardization 118 6.3.4 Pixel Intensities Normalization 118 6.3.5 Harmonization 118 6.3.6 Spacing Resampling 120 6.3.7 Image Enhancement 120 6.3.8 Image Denoising 120 6.3.9 Lowering Dimensionality at the Imaging Level for Deep Learning 121 6.4 Learning Strategies 123 6.4.1 Transfer Learning 123 6.4.2 Multi-task Learning 125 6.4.3 Ensemble Learning 125 6.4.4 Multimodal Learning 125 6.4.5 Federated Learning 128 6.5 Interpretability and Trustworthiness of Artificial Intelligence 129 6.5.1 Reproducibility 130 6.5.2 Traceability 130 6.5.3 Explainability 131 6.5.4 Trustworthiness 131 References 132 7 Data Preparation for AI Analysis 137 7.1 Introduction 137 7.2 Data Quality and Numerosity 139 7.2.1 Intrinsic Image Quality 139 7.2.2 Image Diagnostic Quality 140 7.2.3 Image Quality for AI Analyses 141 7.3 Data Preprocessing for Machine Learning Analyses 144 7.3.1 The Machine Learning Pipeline 147 7.3.2 The Machine Learning Pipeline: A Case Study 148 References 150 8 Current Applications of AI in Medical Imaging 155 8.1 Introduction 155 8.2 Detection 157 8.3 Classification 158 8.4 Segmentation 159 8.4.1 Monitoring 161 8.4.2 Prediction 161 8.4.3 Additional Applications 162 8.4.3.1 Image Enhancement and Reconstruction 162 8.4.4 Workload Reduction? 162 8.5 Conclusions 163 References 164