Diabetic Retinopathy Screening Case Study

This case study explores the development of an assurance case for an AI system designed to detect diabetic retinopathy from retinal images.

Background

Diabetic retinopathy is a leading cause of blindness worldwide. Early detection through regular screening can prevent vision loss. AI systems are increasingly being deployed to assist healthcare professionals in analysing retinal images for signs of the disease.

The System

The hypothetical system in this case study:

• Analyses retinal fundus photographs
• Classifies images by severity of diabetic retinopathy
• Provides recommendations for referral to specialists
• Integrates with clinical workflows

Assurance Goals

The primary assurance goal for this system might be:

“The AI-assisted diabetic retinopathy screening system is safe and effective for use in NHS primary care settings.”

Key Considerations

Safety

• False negative rates and their clinical implications
• Integration with existing clinical pathways
• Human oversight and final decision-making

Effectiveness

• Performance across different patient populations
• Comparison with human expert performance
• Real-world deployment performance vs. clinical trial results

Fairness

• Performance across demographic groups
• Access and availability considerations
• Training data representativeness

Transparency

• Explainability of AI decisions for clinicians
• Patient communication about AI involvement
• Documentation of system limitations

Discussion Questions

1. What evidence would you need to support a claim about the system’s safety?
2. How would you address potential biases in the training data?
3. What stakeholders should be involved in developing this assurance case?
4. How should the assurance case be updated as the system is deployed and used?

Further Reading

• Standards and their Role in Assurance
• DCB0129/DCB0160 Clinical Risk Management Standards