Diabetic retinopathy screening with confocal fundus camera and artificial intelligence - assisted grading

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) screening by ophthalmologists is resource-intensive.
• Automated DR detection using Artificial Intelligence (AI) presents a potential clinical and economic alternative.
• Evaluating AI performance in real-world clinical settings is crucial for adoption.

Purpose of the Study:

• To assess the performance of a confocal fundus imaging system (DRSplus) combined with an AI algorithm (RetCAD) for automated DR detection.
• To compare AI-based DR grading against ophthalmologist grading in a real-world clinical environment.
• To determine the clinical and economic viability of AI-assisted DR screening.

Main Methods:

• A total of 506 patients with diabetes underwent retinal imaging using the DRSplus system.
• Retinal images were graded by both an ophthalmologist and the RetCAD AI algorithm using the International Clinical Diabetic Retinopathy severity scale.
• Performance metrics, including sensitivity and specificity with 95% Confidence Intervals, were calculated at both eye-level and patient-level.

Main Results:

• The AI algorithm achieved high performance in detecting referable diabetic retinopathy.
• Eye-level sensitivity was 97.18% and specificity was 93.73%.
• Patient-level sensitivity reached 98.70% and specificity was 91.06%, with only 2.58% of eyes deemed ungradable by the AI.

Conclusions:

• The combination of DRSplus and RetCAD AI is a dependable solution for real-world DR screening.
• The system's high sensitivity ensures that patients needing further evaluation are appropriately referred.
• AI-assisted DR screening offers a promising, efficient alternative to traditional ophthalmologist-based screening.