Back to all news
UK Researchers Launch Unbiased AI Testing Platform for Diabetic Retinopathy Screening
Researchers unveiled a new independent platform for fair, large-scale evaluation of commercial AI algorithms detecting diabetic eye disease within the NHS.
Key Details
- 1First real-world head-to-head platform to evaluate commercial AI algorithms for NHS use in detecting diabetic eye disease.
- 2Eight CE marked AI algorithms tested on 1.2 million retinal images from one of the UK’s largest and most diverse screening programs.
- 3Algorithm performance (accuracy) ranged from 83.7% to 98.7% for clinically relevant diabetic eye disease, outperforming or matching human graders.
- 4Algorithms’ accuracy remained consistent across varied ethnic groups; first such assessment at this scale.
- 5AI analyzed images in 240ms–45s compared to up to 20 minutes for humans, vastly improving efficiency.
- 6Funded by NHS Transformation Directorate, The Health Foundation, and Wellcome Trust; study published in The Lancet Digital Health.
Why It Matters
This platform addresses critical gaps in algorithmic fairness, transparency, and efficiency for imaging AI, setting a rigorous new standard for evaluation and deployment in national health systems. Its scalable, unbiased testing approach could serve as a blueprint for clinical AI adoption across radiology and other disease areas.
Source
EurekAlert
Related News
•EurekAlert
NYU AI Model Improves 5-Year Breast Cancer Risk Prediction via 3D Mammograms
NYU researchers developed an AI model using longitudinal 3D mammograms that outperforms single-scan and 2D-based tools in predicting 5-year breast cancer risk.
•EurekAlert
Explainable AI Predicts High-Risk Heart Attack Patients for Imaging and Treatment
Researchers developed an explainable AI scoring system to predict intramyocardial hemorrhage risk after severe heart attacks, potentially guiding imaging decisions.
•EurekAlert
AI Framework Accelerates Aortic Aneurysm Risk Prediction from Imaging
Researchers developed BioPINN-LM, combining physics-informed neural networks and multimodal large language models to deliver fast, interpretable risk assessments for ascending thoracic aortic aneurysms.