Related Experiment Video
Updated: Jun 25, 2026
09:17
Using Retinal Imaging to Study Dementia
Published on: November 6, 2017
21.6K
Screening for diabetic retinopathy with artificial intelligence: a real world evaluation
Silvia Burlina1, Sandra Radin2, Marzia Poggiato2
1Diabetes and Endocrinology Unit, ULSS8 Berica, Arzignano, Veneto, VI, Italy. silvia.burlina@aulss8.veneto.it.
Acta Diabetologica
|July 12, 2024
Summary
The DAIRET artificial intelligence system demonstrated high sensitivity in detecting referable diabetic retinopathy (DR) but had lower specificity due to false positives, impacting its cost-effectiveness for DR screening.
Area of Science:
- Ophthalmology
- Medical Artificial Intelligence
- Diabetic Retinopathy Screening
Background:
- Diabetic retinopathy (DR) screening is crucial for preventing blindness in diabetic patients.
- Artificial intelligence (AI) offers potential for enhancing DR screening efficiency.
Purpose of the Study:
- To compare the real-world performance of the DAIRET AI system against ophthalmologists for diabetic retinopathy detection.
Main Methods:
- Retinal fundus images were acquired from 958 diabetic patients using a nonmydriatic camera.
- Images were analyzed by the DAIRET AI system and independently by an ophthalmologist.
- Performance was compared between the AI and human grader.
Main Results:
- The DAIRET system achieved 100% sensitivity for moderate or worse DR and 84% sensitivity for mild DR.
- Specificity for detecting the absence of DR was 59%, limited by a high false-positive rate.
- 867 patients (90.5%) had images suitable for evaluation.
Conclusions:
- DAIRET shows excellent sensitivity for identifying referable diabetic retinopathy compared to human graders.
- The system's low specificity due to false positives may limit its clinical cost-effectiveness.
