Automated diabetic retinopathy detection with two different retinal imaging devices using artificial intelligence: a...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss.
• Early detection and management of DR are crucial to prevent blindness.
• Automated algorithms offer potential for efficient DR screening.

Purpose of the Study:

• To evaluate the diagnostic performance of an automated artificial intelligence (AI) algorithm for diabetic retinopathy detection.
• To compare the AI algorithm's performance using two different retinal imaging systems: a conventional fundus camera and a white LED confocal scanner.

Main Methods:

• 165 diabetic subjects (330 eyes) underwent retinal imaging using both camera types on the same day.
• Images were analyzed for referable diabetic retinopathy (RDR) by retina specialists and the EyeArt AI software.
• Sensitivity, specificity, and AUC were calculated to assess diagnostic performance.

Main Results:

• The AI algorithm achieved 90.8% sensitivity and 75.3% specificity with the conventional fundus camera.
• With the white LED confocal scanner, the AI achieved 94.1% sensitivity and 86.8% specificity.
• The area under the curve (AUC) was significantly higher for the white LED confocal scanner (p=0.0023).

Conclusions:

• Automated AI image analysis software is adaptable to various imaging technologies.
• The AI algorithm demonstrated superior diagnostic performance with the white LED confocal scanner.
• Further validation in large-scale screening programs is recommended.