Artificial Intelligence in Radiation Oncology

Contents
Preface
List of Contributors
Part 1 Define the Future
Chapter 1 Clinical Radiation Oncology in 2040: Vision for Future Radiation Oncology from the Clinical Perspective
1. The Diagnosis
2. In 2040, AI will Facilitate Anticipatory Predictions to Improve Early Detection
3. In 2040, AI-based Approaches will Improve Consistency and Accessibility of High Quality Pathology Interpretation, and Provide Prognostic and Predictive Functionality with Greater Randomized Supporting Data
4. In 2040, Advanced Statistical Approaches will Drive or Provide Critical Characteristics in Staging
5. In 2040, AI will Provide Improved Decision Support to Improve Appropriate Personalized Treatment, and Drive Targeted Areas in Patient-Centered Goals of Care
6. In 2040, Algorithmic Tools Built into EHR Systems will Ease the Process of Record Review, Ordering, and Documentation
6.1. Radiotherapy planning
7. In the Year 2040, AI-Assisted Segmentation and Treatment Planning will be Commonplace and Improve the Quality and Accessibility of Radiotherapy Plans, with Particular Impacts on Access to Care and Active Replanning
7.1. Clinical management
8. In 2040, AI will Enable the Use of Multiple Sources of Data to Help Physicians Better Manage Patient Quality of Life
References
Chapter 2 A Vision for Radiation Oncology in 2030
1. Introduction
1.1. The growing cancer burden
1.2. The growing data challenge
2. A Vision for the Future
2.1. Automation
2.2. Augmentation
2.3. Amplification
3. The Radiation Oncology AI Toolkit
3.1. Training and learning paradigms
3.2. Toolkit selection
4. The Trajectory of AI Tools into 2030
5. Summary
References
Part 2 Strategy
Chapter 3 Lessons from Artificial Intelligence Applications in Radiology for Radiation Oncology
1. Evolution of the Artificial Intelligence Ecosystem
1.1. Expectations and disappointments of AI
1.2. Technical ecosystem helping research and development in AI
2. Different Types of Machine Learning Tools
3. Digital Transformation of Radiology and Computer-Aided Diagnosis (CAD)
3.1. Computer-aided diagnosis
3.2. Technical lesson learned from radiology AI
3.2.1. Limitations of CNNs
3.2.2. Data quality and pre-processing
3.2.3. Data volume and data mix
3.2.4. Expert labeling and curation
3.3. Lessons from poor clinical adoption of AI tools in radiology
4. Comparing Radiology and Radiation Oncology
5. Quantitative Imaging; For New Insights
6. Productivity Improvement and Workflow Optimization
7. Integration of AI Tools into Workflow and New Intelligence Management System (IMS)
8. Ten Principles for Good Machine Learning Practice
9. Conclusion
Acknowledgments
References
Chapter 4 Open Access Data to Enable AI Applications in Radiation Therapy
1. Introduction
2. AI in Radiation Therapy and Cancer Imaging
3. Open Access Data Repositories
4. Distributed ML Without Data Sharing
5. Acquisition, Curation, and Quality
6. Annotations and Labeled Data
7. Remaining Challenges
8. Conclusions
References
Part 3 AI Tools
Chapter 5 Science and Tools of Radiomics for Radiation Oncology
1. Radiomics Definition and History
2. Input Data and Preprocessing
3. Segmentation
4. Radiomic Feature Extraction and Standardization
5. Software Tools for Radiomic Feature Extraction
6. Feature Selection and Dimensionality Reduction
7. Radiomic Analyses
8. Conclusions
References
Chapter 6 Proposed Title: Artificial Intelligence for Image Segmentation in Radiation Oncology
1. Importance of Segmentation in Radiation Oncology
2. Review of Deep Learning Technologies in Medical Image Segmentation
2.1. Fully Convolution Network (FCN)
2.2. U-Net
2.3. Generative Adversarial Network (GAN)
3. Evaluation of Auto-segmentation Performance
4. Challenges in Adopting AI Segmentation in Clinical Practice
4.1. Generalization error
4.2. Variation in contouring standard
5. Summary
References
Chapter 7 Knowledge Representation for Radiation Oncology
1. Introduction
2. Knowledge Representation
3. Unified Medical Language System and Constituents
4. UMLS Limitations
5. KR Outside UMLS
6. Medical NLP Backed by KR
7. Applications of Knowledge Representation
8. Machine Learning Backed by KR
9. ANNs, Knowledge Representation, and Explainability
10. The Future of KR in Medicine
11. Conclusion
References
Chapter 8 Natural Language Processing for Radiation Oncology
1. Introduction
2. What Is NLP?
2.1. The history of NLP
2.2. NLP definitions
2.3. NLP transformation and representation methods
3. NLP in Medicine
3.1. The linguistic string project
3.2. The realtime outbreak and disease surveillance system
3.3. Predicting patient outcomes
3.4. Monitoring adverse drug events
3.5. Processing medical literature for clinician use
3.6. Design and implementation of clinical trials
3.7. Future applications/directions
4. NLP in Oncology
4.1. Radiographic surveillance and diagnosis
4.2. Detailed pathological, molecular, and genomic features
4.3. Identifying patient cohorts from EMR
4.4. Identifying cancer stage from EMR
4.5. Risk assessment
4.6. Clinical outcomes
4.7. Identifying social determinants of care and identifying healthcare gaps
4.8. Summary
5. NLP in Radiation Oncology
5.1. Big data analysis
5.2. Understanding complex radiation histories
5.3. Overcoming non-standardized nomenclature
5.4. Improving documentation and communication of radiation histories
5.5. Treatment-related toxicity
5.6. Real-time management
6. Conclusion
References
Part 4 AI Applications
Chapter 9 Knowledge-Based Treatment Planning: An Efficient and Reliable Planning Technique towards Treatment Planning Automation
1. Introduction
2. Knowledge-Based Treatment Planning (KBP)
2.1. Library-based approach
2.2. Overlap Volume Histogram (OVH)
2.3. Correlation between OVH and DVH
2.4. Selection of DVH values as the optimization objectives for a new patient
3. Applications of KBP
3.1. KBP in head and neck cancer
3.2. KBP in prostate cancer
3.3. KBP in lung, breast and GI cancers
3.4. Cross-institutional application of KBP
3.5. Multi-modality application of KBP
4. Summary
References
Chapter 10 Artificial Intelligence in Radiation Therapy Treatment Planning
1. Introduction
2. Radiation Therapy Treatment Planning Workflow and Automation
2.1. Automated Rule Implementation and Reasoning (ARIR)
2.2. Knowledge-based Planning (KBP) for Dose Volume Histogram (DVH) prediction
2.3. Multicriteria Optimization (MCO)
2.4. Voxel dose prediction and hard-AI
3. AI Approaches in Radiation Therapy Treatment Planning
3.1. Manual Feature Extraction in Dose Volume Histogram (DVH)-based Knowledge-based Planning (KBP)
3.2. Automatic feature extraction: Convolutional Neural Network (CNN)
3.3. Generative Adversarial Neural network (GAN)
3.4. Reinforcement Learning (RL): Optimize machine parameters
4. Other AI Methods and Considerations
Acknowledgment
References
Chapter 11 Clinical Application of AI for Radiation Therapy Treatment Planning
1. Introduction
2. Problem Definition, Scope, and Data Curation
3. Technical Validation
4. Clinical Validation
5. Clinician Trust, AI Explainability, and Bias
6. Special Considerations for AI Treatment Planning Implementation
7. Quality Assurance, Re-Training, and Maintenance
8. Summary
References
Chapter 12 Using AI to Predict Radiotherapy Toxicity Risk Based on Patient Germline Genotyping
1. Introduction
2. Machine Learning Approaches to GWAS
2.1. Statistical analysis approaches
2.1.1. Multiple hypothesis correction
2.1.2. Population structure correction
2.1.3. Genotype imputation
2.1.4. Fine mapping
2.2. Machine learning approaches
2.2.1. Selecting genomic features as inputs to model building
2.2.2. Validating machine learning models
2.2.3. Methods for developing prediction signatures from GWAS
2.2.3.1. Support vector machines
2.2.3.2. Penalized logistic regression
2.2.3.3. Random forest models
2.2.3.4. Deep learning
2.2.3.5. Network analysis
2.3. A hybrid method of machine learning and statistical analysis
2.4. Integration of dose-volume and genetic factors into NTCP models
3. The Identification of Biological Correlates Associated with Toxicity: A Key Advantage of Machine Learning Signatures in GWAS
4. Conclusion
Acknowledgments
References
Chapter 13 Utilization of Radiomics in Prognostication and Treatment Response
1. Radiomics as an Adjunct to Prognosis
2. Prediction of Treatment Response
3. Tracking Treatment Response
4. Radiomics as a Surrogate for Pathologic Information
5. Artificial Intelligence in the Development of Clinically Focused Radiomics Signatures
6. Conclusions and Future Directions
References
Chapter 14 How AI can Help us Understand and Mitigate Error Propagation in Radiation Oncology
1. Introduction
2. Patient Safety in Radiation Oncology
3. The Radiation Oncology Workflow: Constructing a Reference Timeline
4. Creating a Prototype Statistical Model
5. Understanding Error Reporting Data in Our System:Past and Present
6. Understanding Error Reporting Data: What’s in the Future?
7. Envisioning an AI-powered Error Mitigation System
8. The Challenges and Pitfalls of AI
9. Conclusion
References
Part 5 Assessment and Outcomes
Chapter 15 Ethics and Artificial Intelligence in Radiation Oncology
1. Ethical Principles in Medicine
1.1. Respect for autonomy
1.2. Beneficence and nonmaleficence
1.3. Justice
2. Ethical Concerns for AI in Medicine and Radiation Oncology
2.1. Inconclusive evidence
2.2. Inscrutable evidence
2.3. Misguided evidence
2.4. Unfair outcomes
2.5. Transformative effects
2.6. Traceability
3. Emerging AI Tools in Radiation Oncology
3.1. Adoption of AI tools with potential biases
3.2. AI in radiation oncology and global health
4. Ethics and the Future of AI in Radiation Oncology
Acknowledgments
References
Chapter 16 Evaluation of Artificial Intelligence in Radiation Oncology
1. Overview of AI in Radiation Oncology
2. A Framework for the Evaluation of AI in Healthcare
3. AI Evaluations in Radiation Oncology
4. Conclusions
References
Index