Publicaciones › Inteligencia artificial y análisis automático de imágenes
Performance of AI-based diabetic retinopathy screening is highly dependent on evaluation setting: a five-year, multi-framework study
El rendimiento del cribado de retinopatía diabética basado en inteligencia artificial depende en gran medida del entorno de evaluación: un estudio de cinco años y múltiples marcos
| Año | 2026 |
|---|---|
| Fuente | Scientific Reports |
| Tipo | article |
| Autores (14) | Gwenolé Quellec, Mathieu Lamard, Sarah Christina Matta, Laurent Borderie, Philippe Zhang, Paola Ortega Saborio, Pierre Deman, Alexandre Le Guilcher, Ali Erginay, Anna‐Maria Kubin, Nina Maria Hautala, Petri Huhtinen, Béatrice Cochener, Pascale Massin |
| Posición de Erginay | 9 de 14 |
| Citas | 0 (OpenAlex) |
| Enlaces | DOI 10.1038/s41598-026-70676-2 · OpenAlex · Texto completo en acceso abierto |
| Líneas | Inteligencia artificial y análisis automático de imágenes Cribado, telemedicina (OPHDIAT) y epidemiología de la retinopatía diabética Edema macular diabético y sus tratamientos Retinopatía diabética: clínica, función y progresión Imagen retiniana: OCT, OCT-angiografía y ultra-gran campo |
En pocas palabras
Se estudió el rendimiento de un sistema de inteligencia artificial para detectar retinopatía diabética y edema macular en fotografías de fondo de ojo. El estudio se llevó a cabo en diferentes entornos y se encontró que el rendimiento del sistema variaba significativamente dependiendo del entorno de evaluación. Esto sugiere que el rendimiento del sistema no es una propiedad fija, sino que depende del contexto en el que se evalúa.
Resumen divulgativo generado automáticamente a partir del resumen original; puede contener imprecisiones.
Resumen original
Abstract Despite near-perfect performance reported on benchmark datasets, the real-world behavior of artificial intelligence (AI) systems for diabetic retinopathy (DR) screening remains insufficiently characterized. Here, we report a multi-framework evaluation of OphtAI, a CE-marked AI system for automated detection of referable DR and diabetic macular edema from color fundus photographs. The system was initially validated on the Messidor-2 benchmark dataset and subsequently assessed across three independent evaluation settings: a large-scale masked comparative study (US Veterans Affairs), an external validation using handheld fundus photography in Finland, and an open, large-scale comparative evaluation within the UK National Health Service. Across these evaluation frameworks, substantial variability in observed sensitivity and specificity was found. These variations occurred across settings differing in imaging devices, population characteristics, referral definitions, and handling of ungradable images. In masked settings, interpretation of comparative performance was limited by the lack of system-level attribution, while open evaluations remained sensitive to protocol design. Taken together, these results show that performance in AI-based DR screening is not a fixed property of a system, but an emergent property of the evaluation framework. This work calls for context-aware validation strategies and more standardized evaluation protocols for clinical AI systems.