Contribution to a book

Front Cover -- Trustworthy AI in Medical Imaging -- Copyright -- Contents -- List of contributors -- Preface -- 1 Preliminaries -- 1 Introduction to trustworthy AI for medical imaging -- 1.1 Lecture plan -- References -- 2 The fundamentals of AI ethics in medical imaging -- 2.1 Introduction -- 2.2 From ethics to AI ethics, to responsible and trustworthy AI -- 2.2.1 AI regulation -- 2.3 The need for trustworthy AI in medical imaging -- 2.3.1 Robustness -- 2.3.2 Validation, transparency, and reproducibility -- 2.3.3 Bias &amp -- fairness -- 2.3.4 Explainability, interpretability, and causality -- 2.3.5 Privacy &amp -- data protection -- 2.3.6 Collaborative learning -- 2.4 Addressing ethical challenges in practice -- 2.4.1 Operationalizing AI ethics -- 2.4.2 Teaching AI ethics -- 2.4.2.1 Integration of AI ethics in computer science curricula -- 2.4.2.2 AI ethics in medical curricula -- 2.4.3 Stakeholder engagement -- 2.5 Conclusion -- References -- 2 Robustness -- 3 Machine learning robustness: a primer -- 3.1 Definition -- 3.1.1 Robustness complements (i.i.d.) generalizability -- 3.1.2 Robustness is a requirement of trustworthy AI -- 3.1.3 Adversarial vs nonadversarial robustness -- 3.1.4 Robustness measurement -- 3.1.5 Robustness indicators: reproducibility and explainability -- 3.2 Challenges -- 3.2.1 Data bias: train-serving skew -- 3.2.2 Model complexity: a double-edge sword -- 3.2.3 Underspecified ML pipeline: one pipeline, many models -- 3.3 Robustness assessment -- 3.3.1 Nonadversarial shifts: from synthetic to real-world -- 3.3.2 Adversarial attacks: categories and aims -- 3.3.2.1 White-box adversarial attacks -- 3.3.2.2 Black-box adversarial attacks -- 3.3.2.3 Physical adversarial attacks -- 3.3.3 DL software testing: test-driven model verification -- 3.3.3.1 Pseudo oracle -- 3.3.3.2 Test adequacy criteria