Comparative Analysis of AI and Ophthalmologist Grading in Diabetic Retinopathy Detection

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) presents a significant global health challenge requiring scalable screening solutions.
• Current teleophthalmology methods face limitations in addressing the growing need for DR screening.
• Artificial intelligence (AI) offers potential for efficient and accurate DR detection.

Purpose of the Study:

• To evaluate the diagnostic performance of the AI DRISTi system (Version 2.1).
• To compare AI grading against ophthalmologist grading for diabetic retinopathy (DR) and diabetic macular edema (DME).
• To assess the AI system's accuracy across various fundus camera platforms.

Main Methods:

• Retrospective, observational, non-interventional validation study.
• Comparison of AI DRISTi system output with ophthalmologist grading.
• Analysis of 739 color fundus images from multiple camera types (Topcon, CrystalVue, Canon, Zeiss).

Main Results:

• The AI system achieved high accuracy: 93.36% for mtmDR, 98.64% for vtDR, and 97.97% for DME.
• Sensitivity and specificity were robust across all evaluated camera types.
• Performance metrics included sensitivity (95.03% for mtmDR, 96.92% for vtDR, 92.85% for DME) and specificity (92.90% for mtmDR, 99.01% for vtDR, 98.88% for DME).

Conclusions:

• The AI DRISTi system (Version 2.1) exhibits strong diagnostic performance for mtmDR, vtDR, and DME.
• AI system performance is comparable to leading commercial AI systems and consistent across multiple camera platforms.
• The AI system shows promise as an adjunctive tool for large-scale DR screening, aiding early detection and preventing vision loss.