Case Studies
Explore real-world examples of assurance cases across different domains. These case studies demonstrate how the TEA methodology can be applied to various AI and data-driven systems.
Available Case Studies
| Case Study | Domain | Assurance Goal |
|---|---|---|
| Explainable Diabetic Retinopathy Screening System | Healthcare | Explainability |
| Fair Crop Damage Assessment System | Agriculture | Fairness |
| Equitable Flood Risk Assessment System | Environmental | Fairness |
| Explainable Student Learning Assessment System | Education | Explainability |
| Equitable Personalised Pharmaceutical Formulation System | Pharmaceutical | Fairness |
| Transparent Clinical GenAI System with Legacy Data | Healthcare | Transparency |
| Explainable Reinforcement Learning Agent for Air Traffic Control | Aviation | Explainability |
| Safe Adaptive Allocation in a Bayesian Platform Clinical Trial | Healthcare | Safety |
| Balancing Privacy and Utility in Census Disclosure Control | Public Sector | Privacy |
| Equitable Identification in Aerial Facial Recognition | Security and Defence | Fairness |
Healthcare
- Explainable Diabetic Retinopathy Screening System - A case study exploring assurance for AI-powered diabetic retinopathy screening with a focus on explainability
- Transparent Clinical GenAI System with Legacy Data - A case study exploring transparency and accountability for a GenAI clinical decision support system trained on patient data collected under historical consent frameworks
- Safe Adaptive Allocation in a Bayesian Platform Clinical Trial - A case study exploring how to assure the safety, and supporting explainability, of a Bayesian response-adaptive randomisation engine that updates patient allocation mid-trial in a multi-arm platform trial
Agriculture
- Fair Crop Damage Assessment System - A case study exploring fairness assurance for AI-powered agricultural damage assessment
Environmental
- Equitable Flood Risk Assessment System - A case study exploring fairness assurance for AI-powered flood risk prediction and resource allocation
Education
- Explainable Student Learning Assessment System - A case study exploring explainability assurance for AI-powered educational assessment and learning analytics
Pharmaceutical
- Equitable Personalised Pharmaceutical Formulation System - A case study exploring fairness assurance for AI-powered personalised drug formulations trained on demographically biased data
Aviation
- Explainable Reinforcement Learning Agent for Air Traffic Control - A case study exploring how to assure the explainability of a reinforcement learning agent undergoing a supervised operational trial as a decision-support tool for en route air traffic controllers
Public Sector
- Balancing Privacy and Utility in Census Disclosure Control - A case study exploring how to assure the privacy–utility trade-off of a differentially private disclosure-control system used to publish national census statistics
Security and Defence
- Equitable Identification in Aerial Facial Recognition - A case study exploring how to assure demographic fairness in a drone-based facial-recognition framework that reconstructs non-frontal faces with a generative AI frontalisation module before identification on energy-efficient neuromorphic classifiers
How to Use These Case Studies
Each case study includes:
- Overview - Background on the domain and system being assessed
- System Description - Technical details of the AI system
- Stakeholders - Key parties with interests in the system’s assurance
- Regulatory Context - Relevant regulations and standards
- Assurance Considerations - Specific concerns for the assurance goal (e.g., fairness, explainability, transparency)
- Deliberative Prompts - Questions for reflection and discussion
- Suggested Strategies - Approaches for developing the assurance case
- Recommended Techniques - Links to relevant TEA Techniques for gathering evidence
These case studies can be used for:
- Self-study - Work through examples at your own pace
- Workshops - Group activities and discussions
- Templates - Starting points for your own assurance cases