Machine Learning in Clinical Neuroscience: Foundations and Applications (Acta Neurochirurgica Supplement Book 134)

This book bridges the gap between data scientists and clinicians by introducing all relevant aspects of machine learning in an accessible way, and will certainly foster new and serendipitous applications of machine learning in the clinical neurosciences. Building from the ground up by communicating the foundational knowledge and intuitions first before progressing to more advanced and specific topics, the book is well-suited even for clinicians without prior machine learning experience. Authored by a wide array of experienced global machine learning groups, the book is aimed at clinicians who are interested in mastering the basics of machine learning and who wish to get started with their own machine learning research. The volume is structured in two major parts: The first uniquely introduces all major concepts in clinical machine learning from the ground up, and includes step-by-step instructions on how to correctly develop and validate clinical prediction models. It also includes methodological and conceptual foundations of other applications of machine learning in clinical neuroscience, such as applications of machine learning to neuroimaging, natural language processing, and time series analysis. The second part provides an overview of some state-of-the-art applications of these methodologies. The Machine Intelligence in Clinical Neuroscience (MICN) Laboratory at the Department of Neurosurgery of the University Hospital Zurich studies clinical applications of machine intelligence to improve patient care in clinical neuroscience. The group focuses on diagnostic, prognostic and predictive analytics that aid in decision-making by increasing objectivity and transparency to patients. Other major interests of our group members are in medical imaging, and intraoperative applications of machine vision. Contents 1: Machine Intelligence in Clinical Neuroscience: Taming the Unchained Prometheus 1.1 Preface References Part I: Clinical Prediction Modeling 2: Foundations of Machine Learning-Based Clinical Prediction Modeling: Part I—Introduction and General Principles 2.1 Introduction 2.2 Machine Learning: Definitions 2.3 Optimization: The Central Dogma of Learning Techniques 2.4 Explanatory Modeling Versus Predictive Modeling 2.5 Workflow for Predictive Modeling 2.6 Conclusion References 3: Foundations of Machine Learning-Based Clinical Prediction Modeling: Part II—Generalization and Overfitting 3.1 Introduction 3.2 Overfitting The Bias-Variance Trade-Off Combatting Overfitting: Resampling Considerations on Algorithm Complexity Data Leakage 3.3 Importance of External Validation in Clinical Prediction Modeling 3.4 Feature Reduction and Selection 3.5 Conclusion References 4: Foundations of Machine Learning-Based Clinical Prediction Modeling: Part III—Model Evaluation and Other Points of Significance 4.1 Introduction 4.2 Evaluation of Classification Models The Importance of Discrimination and Calibration Model Discrimination Area Under the Curve (AUC) Accuracy Sensitivity and Specificity Positive Predictive Value (PPV) and Negative Predictive Value (NPV) F1 Score Model Calibration Calibration Intercept and Slope Brier Score Other Calibration Metrics Recalibration Techniques 4.3 Evaluation of Regression Models 4.4 Points of Significance Choosing a Cutoff for Binary Classification Sample Size Standardization One-Hot Encoding Missing Data and Imputation Class Imbalance Extrapolation 4.5 Conclusion References 5: Foundations of Machine Learning-Based Clinical Prediction Modeling: Part IV—A Practical Approach to Binary Classification Problems 5.1 Introduction 5.2 Setup and Pre-processing Data R Setup and Package Installation Importing Data Check the Imported Data Reformat Categorical Variables Remove Unnecessary Columns Enable Multicore Processing Partition the Data for Training and Testing Impute Missing Data Variable Selection Using Recursive Feature Elimination Get a Final Overview of the Data 5.3 Model Training Setting Up the Training Structure Model Training 5.4 Model Evaluation and Selection Model Training Evaluation Select the Final Model Internal Validation on the Test Set 5.5 Reporting and Visualization Compiling Training Performance Compiling Internal Validation Performance Assessing Variable Importance 5.6 Conclusion References 6: Foundations of Machine Learning-Based Clinical Prediction Modeling: Part V—A Practical Approach to Regression Problems 6.1 Introduction 6.2 Setup and Pre-processing Data Reformat Categorical Variables Remove Unnecessary Columns Enable Multicore Processing Partition the Data for Training and Testing Impute Missing Data Variable Selection using Recursive Feature Elimination Get a Final Overview of the Data 6.3 Model Training Setting Up the Training Structure Model Training 6.4 Model Evaluation and Selection Model Training Evaluation Select the Final Model Internal Validation on the Test Set 6.5 Reporting and Visualization Compiling Training Performance Compiling Internal Validation Performance Assessing Variable Importance 6.6 Conclusion References 7: Foundations of Feature Selection in Clinical Prediction Modeling 7.1 Introduction 7.2 Foundations of Feature Selection 7.3 Statistical Filtering Methods Correlation and Significance Testing 7.4 Algorithmic Wrapper Methods Feature Importance-Based Purposeful Variable Selection Algorithm Recursive Feature Elimination 7.5 Intrinsic Methods Tree- and Rule-Based Methods Lasso 7.6 Unsupervised Feature Selection Methods 7.7 Conclusions References 8: Dimensionality Reduction: Foundations and Applications in Clinical Neuroscience 8.1 Introduction 8.2 Feature Engineering Using Imaging-Derived Phenotypes (IDPs) 8.3 Dimensionality Reduction Using Principal Component Analysis 8.4 Methodological Pitfalls Scale Invariance The Optimal Number of PCs 8.5 Conclusion References 9: A Discussion of Machine Learning Approaches for Clinical Prediction Modeling 9.1 Introduction 9.2 Early Applications of Machine Learning to Clinical Applications 9.3 Supervised Machine Learning Approaches Regression Analysis Support Vector Machine Decision Trees and Random Forest Artificial Neural Networks Naïve Bayes 9.4 Unsupervised Machine Learning Approaches Clustering 9.5 Conclusion References 10: Foundations of Bayesian Learning in Clinical Neuroscience 10.1 Introduction 10.2 Bayes Theorem 10.3 Bayesian Networks 10.4 Naïve Bayes Classifiers 10.5 Discussion 10.6 Conclusion References 11: Introduction to Deep Learning in Clinical Neuroscience 11.1 Introduction 11.2 Materials and Methods: Useful DL Methods in Clinical Neuroscience Pre-processing of MRI Data Segmentation of Region of Interest (ROI) Deep Convolutional Neural Networks (CNNs) Deep Autoencoders (AEs) Generative Adversarial Networks (GANs) Techniques to Effectively Combining Several Small Datasets 11.3 Results: DL-Assisted Diagnostics in Gliomas Results of Tumor Segmentation Performed by DL Instead of Manual Outline Prediction of Glioma Subtypes of New Patients with MRIs Only Results Following Expanding Training Data by DL Results Following Fitting Data from Several Sources with Significant Variability 11.4 Discussion 11.5 Concluding Remarks References 12: Machine Learning-Based Clustering Analysis: Foundational Concepts, Methods, and Applications 12.1 Introduction 12.2 Connectivity-Based Clustering 12.3 Centroid-Based Clustering 12.4 Density-Based Clustering 12.5 Dimensionality Reduction 12.6 Applications Adult Spinal Deformity Sepsis Common Pitfalls and Proposed Solutions 12.7 Conclusions References 13: Deployment of Clinical Prediction Models: A Practical Guide to Nomograms and Online Calculators 13.1 Introduction 13.2 Nomograms 13.3 Online Calculators 13.4 Other Methods of Deployment 13.5 Discussion 13.6 Conclusion References 14: Updating Clinical Prediction Models: An Illustrative Case Study 14.1 Introduction 14.2 Methods Study Population and Design Model Development Set External Data Set for Domain Updating Outcome Definition Predictor Variables Statistical Analysis Updating Strategies Reference Method Recalibration Method Model Revision Model Extension 14.3 Results 14.4 Discussion References 15: Is My Clinical Prediction Model Clinically Useful? A Primer on Decision Curve Analysis 15.1 Introduction 15.2 Methodology Decision Curves Interventions Avoided 15.3 Example 15.4 Final Comment References Part II: Neuroimaging 16: Introduction to Machine Learning in Neuroimaging 16.1 Introduction 16.2 Main Part Image Preprocessing Dimensionality Reduction Feature Selection fMRI Analyses: Supervised vs. Unsupervised Decoding/Encoding Framework Clustering 16.3 Conclusion References 17: Machine Learning Algorithms in Neuroimaging: An Overview 17.1 Introduction 17.2 The Radiomic Workflow 17.3 Introduction to Deep Learning Algorithms for Imaging Convolutional Neural Networks (CNNs) Architecture Convolution and Kernels Hyperparameter Optimization Activation Function and Backpropagation Backpropagation Optimization and Network Training Pooling, Fully Connected Layers, and Last Activation Function Overfitting and Dropout 2D vs. 3D CNN Transfer Learning Available CNN Architectures Generative Adversarial Networks Data Availability and Privacy Deep Learning-Based Tasks in Imaging Image Reconstruction and Restoration Image Synthesis and Super-Resolution Image Registration Image Segmentation, Classification, and Outcome Prediction 17.4 Conclusions References 18: Machine Learning-Based Radiomics in Neuro-Oncology 18.1 Introduction 18.2 Methodological Foundations 18.3 Recent Implications for Neuro-Oncology 18.4 Automated Tumor Segmentation 18.5 Molecular Phenotyping and Radiogenomics 18.6 Prediction of Clinical Outcome 18.7 Discriminating Radiation Necrosis from Tumor Progression and Primary from Secondary Brain Lesions 18.8 Conclusions References 19: Foundations of Brain Image Segmentation: Pearls and Pitfalls in Segmenting Intracranial Blood on Computed Tomography Images 19.1 Introduction 19.2 Segmentation: What, Why, and How 19.3 Multi-label Segmentation 19.4 Segmentation of Blood Detected in Head CT Scans 19.5 Confounders in Segmenting Blood 19.6 Selecting a Segmentation Software 19.7 Practical Segmentation Tips 19.8 Conclusions References 20: Applying Convolutional Neural Networks to Neuroimaging Classification Tasks: A Practical Guide in Python 20.1 Introduction 20.2 Digital Imaging and Communications in Medicine (DICOM) 20.3 Practical Steps 20.4 Image Preprocessing 20.5 Convolutional Neural Network (CNN) Building and Assessment References 21: Foundations of Lesion Detection Using Machine Learning in Clinical Neuroimaging 21.1 Introduction 21.2 Technical Considerations 21.3 Clinical Applications Introduction to Clinical Applications 21.4 Stroke ASPECTS Large Vessel Occlusion (LVO) Identification of Infarct Core and Tissue at Risk/Penumbra Hemorrhagic Transformation Intracranial Hemorrhage 21.5 Multiple Sclerosis 21.6 Neuro-Oncology 21.7 Epilepsy 21.8 Aneurysms 21.9 Neurodegeneration and Others 21.10 Conclusion References 22: Foundations of Multiparametric Brain Tumour Imaging Characterisation Using Machine Learning 22.1 Introduction 22.2 Methodological Foundations Multiparametric Imaging 22.3 Image Preprocessing 22.4 Region of Interest (ROI) Selection 22.5 Feature Extraction and Multiparametric Analysis Machine Learning Classifiers 22.6 Deep Learning in Brain Tumour Characterisation 22.7 Performance Evaluation of ML Algorithms 22.8 Clinical Applications 22.9 Conclusions References 23: Tackling the Complexity of Lesion-Symptoms Mapping: How to Bridge the Gap Between Data Scientists and Clinicians? 23.1 Introduction 23.2 Clarifying the Problem Lesional Localizationism, Lesional Hodotopism, Functional Localizationism From Behavioural Measurements to Cognitive Processes: Leveraging Multidimensional Scores The Complexity of Lesion-Symptom Mapping 23.3 Data-Driven vs. Model-Based Approaches Data-Driven Approaches Model-Driven (Top-Down) Approaches 23.4 How to Capitalize on Multimodal Longitudinal Single Cases? The Value of Multimodal Longitudinal Single Cases A New Paradigm for Combining Single-Case Analysis with the Predictive Power of Machine Learning 23.5 Conclusion References Part III: Natural Language Processing and Time Series Analysis 24: Natural Language Processing: Practical Applications in Medicine and Investigation of Contextual Autocomplete 24.1 Introduction 24.2 Contextual Autocomplete Literature Review 24.3 Contextual Autocomplete: Technical Toolkit Trie Data Structure BoW Model TF-IDF Encoding Support Vector Machine (SVM) Confusion Matrix for Visualizing Model Accuracy 24.4 Conclusion References 25: Foundations of Time Series Analysis 25.1 Introduction 25.2 Foundational Methods Parametric Methods Nonparametric Methods Clinical Applications 25.3 Conclusions References 26: Overview of Algorithms for Natural Language Processing and Time Series Analyses 26.1 Introduction 26.2 Natural Language Processing Preprocessing N-grams Data Representation Bag of Words One-Hot Encoding Word Embeddings: Neural Network Basics Word Embeddings: Learning an Embedding Matrix Word Embeddings: Implementation Recurrent Neural Networks Gated Recurrent Units Long Short-Term Memory (LSTM) Network Convolutional Neural Networks CNNs Applied to NLP 26.3 Time Series Analysis Preprocessing Neural Networks: Multilayer Perceptron Neural Networks: LSTM Neural Networks: CNNs 26.4 Conclusion References Part IV: Ethical and Historical Aspects of Machine Learning in Medicine 27: A Brief History of Machine Learning in Neurosurgery 27.1 Introduction 27.2 The Evolution of Machine Learning in Neurosurgery 1990s: Early Applications in Neurosurgery 2000s: Refinement and Expansion 2010s: Exponential Growth and Adoption of Machine Learning 27.3 Contemporary and Novel Applications 27.4 Conclusion References 28: Machine Learning and Ethics 28.1 Introduction 28.2 Personal Integrity 28.3 Justice and Investments in Information Technology 28.4 Accountability: Who Decides and What Is the Decision Based On? Values and AI Traceability of Decisions and Recommendations 28.5 Discussion References 29: The Artificial Intelligence Doctor: Considerations for the Clinical Implementation of Ethical AI 29.1 Introduction 29.2 Transparency and Explicability 29.3 Fairness and Bias 29.4 Liability and Legal Implications 29.5 Conclusion References 30: Predictive Analytics in Clinical Practice: Advantages and Disadvantages 30.1 Introduction 30.2 Data Considerations: What to Put into a Predictive Tool? Quantity Versus Quality Theoretical Construct and Empirical Construct Analyzing Available Data or Analyzing Clinical Equipoise 30.3 Interpreting the Model’s Output: An Essential Role for the Clinical Neuroscientist Clinical and Scientific Competencies Clinical Neuroscientist’s Vigilance 30.4 Integrating the Model into the Clinical Workflow: Reporting Is Imperative User Trust Transparency Safe Use and Regulatory Approval 30.5 Concluding Remarks References Part V: Clinical Applications of Machine Learning in Clinical Neuroscience 31: Big Data in the Clinical Neurosciences 31.1 Introduction 31.2 Historical Context Within Neurosurgery 31.3 Evolution of Clinical Neurosurgical Databases 31.4 Future Directions 31.5 Conclusion References 32: Natural Language Processing Applications in the Clinical Neurosciences: A Machine Learning Augmented Systematic Review 32.1 Introduction 32.2 Method Study Identification Inclusion and Exclusion Criteria Data Collection and Extraction Analysis 32.3 Results NLP Application Domains NLP for Patient Cohort Identification NLP for Automated Reporting NLP for Data and Information Extraction NLP for Literature Synthesis NLP for Outcome Prediction NLP Analysis NLP Resources 32.4 Discussion 32.5 Conclusion References 33: Machine Learning in Pituitary Surgery 33.1 Introduction 33.2 Machine Learning Applications in Pituitary Surgery Enhanced Preoperative Lesion Characterization Differential Diagnosis Immunohistochemical Characterization of PA CS Invasion by PA Adenoma Tumor Consistency Surgical Outcome and Complication Prediction Gross Total Resection Intraoperative Cerebrospinal Fluid (CSF) Leak Tumor Recurrence and Endocrinological Remission Hyponatremia Drug Treatment Response Costs Limitations Future Directions 33.3 Conclusions References 34: At the Pulse of Time: Machine Vision in Retinal Videos 34.1 Introduction 34.2 Methods Source Data Pre-Processing Filters for Frames Normalization and Noise Reduction Dealing with Blurry Frames Registration Detection/Enhancement of SVP 34.3 Discussion 34.4 Conclusion References 35: Artificial Intelligence in Adult Spinal Deformity 35.1 Introduction 35.2 Methods 35.3 Results 35.4 Discussion Machine Learning Computer Vision/Augmentation Future 35.5 Conclusion References 36: Machine Learning and Intracranial Aneurysms: From Detection to Outcome Prediction 36.1 Introduction 36.2 Machine Learning Applications in the Management of Patients with Intracranial Aneurysms Aneurysm Detection [18–27] (Table 36.1) Aneurysm Rupture Risk and Stability Prediction [28–32] (Table 36.2) Complications and Outcome Prediction [35–43] (Table 36.3) 36.3 Discussion IA Screening and Detection Rupture Risk and Aneurysm Stability Outcome Prediction Delayed Cerebral Ischemia, Vasospasm, and Shunt-Dependent Hydrocephalus Functional Outcome Prediction Periprocedural Outcome Prediction 36.4 Future Directions 36.5 Conclusions References 37: Clinical Prediction Modeling in Intramedullary Spinal Tumor Surgery 37.1 Introduction 37.2 A Primer on Machine Learning and Predictive Analytics 37.3 Defining Outcome Measures for Intramedullary Spinal Cord Tumors 37.4 Available Sources of Data for Prediction Modeling in IMSCT 37.5 Imaging Features and Biomarkers to Predict Outcomes for IMSCT 37.6 Genetic Biomarkers of IMSCT Ependymoma Astrocytoma Hemangioblastoma 37.7 Genome-Wide Association Studies 37.8 Discovery of Biomarkers and Prediction of Therapeutic Responses 37.9 Conclusion References 38: Radiomic Features Associated with Extent of Resection in Glioma Surgery 38.1 Introduction 38.2 Basic Workflow in Radiomics Image Post-Processing and Tumor Segmentation Radiomic Features Feature Selection and Model Creation 38.3 Applications in Neuro-Oncology 38.4 Features Associated with Extent of Resection in Brain Glioma Future Perspectives 38.5 Conclusions References 39: Machine Learning in Neuro-Oncology, Epilepsy, Alzheimer’s Disease, and Schizophrenia 39.1 Introduction 39.2 Materials and Methods Data Extraction PICOS Outline Search Criteria Inclusion and Exclusion Criteria 39.3 Results Neuro-Oncology Epilepsy Alzheimer’s Disease Schizophrenia 39.4 Discussion Neuro-Oncology Epilepsy Alzheimer’s Disease Schizophrenia 39.5 Conclusions References Preface -- Foundations of machine learning-based clinical prediction modeling - Part I: Introduction and general principles -- Foundations of machine learning-based clinical prediction modeling - Part II: Generalization and Overfitting -- Foundations of machine learning-based clinical prediction modeling - Part III: Evaluation and other points of significance -- Foundations of machine learning-based clinical prediction modeling - Part IV: A practical approach to binary classification problems -- Foundations of machine learning-based clinical prediction modeling - Part V: A practical approach to regression problems -- Supervised and unsupervised learning / clustering -- Introduction to Bayesian Modeling -- Introduction to Deep Learning -- Overview of algorithms for machine-learning based clinical prediction modelling -- Foundations of feature selection in clinical prediction modelling -- Dimensionality reduction: Foundations and applications in clinical neuroscience -- Machine learning-based survival modeling: Foundations and Applications -- Making clinical prediction models available: A brief introduction -- Machine Learning-based Clustering Analysis: Foundational Concepts, Methods, and Applications -- Introduction to Machine Learning in Neuroimaging -- Overview of machine learning algorithms in imaging -- Foundations of classification modeling based on neuroimaging -- Foundations of lesion-symptom mapping using machine learning -- Foundations of Machine Learning-Based Segmentation in Cranial Imaging -- Foundations of lesion detection using machine learning in clinical neuroimaging -- Foundations of multiparametric brain tumor imaging characterization -- Radiomics in clinical neuroscience - Overview -- Radiomic feature extraction: Methodological Foundations -- Complexity and interpretability in machine vision -- Foundations of intraoperative anatomical recognition using machine vision -- Machine Vision Foundations -- Natural Language Processing: Foundations and Applications in Clinical Neuroscience -- Foundations of Time Series Analysis -- Overview of algorithms for natural language processing and time series analysis -- History of machine learning in neurosurgery -- The AI doctor - considerations for AI-based medicine -- Ethics of Machine Learning-Based Predictive Analytics -- Predictive analytics in clinical practice: Pro and contra -- Review of machine vision applications in neuroophtalmology -- Prediction Model -- Prediction Model -- Prediction Model -- Topical Review of machine learning in intracranial aneurysm surgery -- Review of applications of machine learning in neuroimaging -- Prediction Model -- An overview of machine learning applications in the Neurointensive Care Unit -- Prediction Model -- Review of natural language processing in the clinical neurosciences -- Review of big data applications in the clinical neurosciences -- Radiomic features associated with extent of resection in glioma surgery The Book Bridges The Gap Between Computer Scientists And Clinicians By Introducing All Relevant Aspects Of Machine Learning In An Accessible Way, And Will Certainly Foster New And Serendipitous Applications Of Machine Learning In The Clinical Neurosciences.the Machine Intelligence In Clinical Neuroscience (micn) Laboratory At The Department Of Neurosurgery Of The University Hospital Zurich Studies Clinical Applications Of Machine Intelligence To Improve Patient Care In Clinical Neuroscience. The Group Focuses On Diagnostic, Prognostic And Predictive Analytics That Aid In Decision-making By Increasing Objectivity And Transparency To Patients. These Algorithms Provide Patients With Objective Information That May Aid In Risk-benefit Discussion, Help Prevent Adverse Events, Improve Outcome, And Patient Safety Overall. Other Major Interests Of Our Group Members Are In Medical Imaging, And Intraoperative Applications Of Machine Vision.the Volume Is Structured In Two Major Parts: The First Uniquely Introduces All Major Concepts In Clinical Machine Learning From The Ground Up, And Includes Step-by-step Instructions On How To Correctly Develop And Validate Clinical Prediction Models. It Also Includes Methodological And Conceptual Foundations Of Other Applications Of Machine Learning In Clinical Neuroscience, Such As Applications Of Machine Learning To Neuroimaging, Natural Language Processing, And Time Series Analysis. The Second Part Provides An Overview Of Some State-of-the-art Applications Of These Methodologies.this Work – Authored By A Wide Array Of Experienced Global Machine Learning Groups – Is Aimed At Clinicians Who Are Interested In Mastering The Basics Of Machine Learning And Who Wish To Get Started With Their Own Machine Learning Research. This book bridges the gap between data scientists and clinicians by introducing all relevant aspects of machine learning in an accessible way, and will certainly foster new and serendipitous applications of machine learning in the clinical neurosciences. Building from the ground up by communicating the foundational knowledge and intuitions first before progressing to more advanced and specific topics, the book is well-suited even for clinicians without prior machine learning experience. Authored by a wide array of experienced global machine learning groups, the book is aimed at clinicians who are interested in mastering the basics of machine learning and who wish to get started with their own machine learning research. The __Machine Intelligence in Clinical Neuroscience (MICN) Laboratory__ at the Department of Neurosurgery of the University Hospital Zurich studies clinical applications of machine intelligence to improve patient care in clinical neuroscience. The group focuses on diagnostic, prognostic and predictive analytics that aid in decision-making by increasing objectivity and transparency to patients. Other major interests of our group members are in medical imaging, and intraoperative applications of machine vision.