Accuracy of Integrated Artificial Intelligence Grading Using Handheld Retinal Imaging in a Community Diabetic Eye...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) and diabetic macular edema (DME) are leading causes of vision loss in diabetic patients.
• Early detection and management are crucial to prevent vision-threatening complications.
• Current screening methods often rely on centralized reading centers (RCs), which can cause delays in diagnosis and treatment.

Purpose of the Study:

• To evaluate the performance of point-of-care (POC) artificial intelligence (AI) in identifying diabetic retinopathy (DR) and diabetic macular edema (DME) using handheld retinal images.
• To compare POC AI assessment with traditional retinal image grading by RCs.

Main Methods:

• A prospective, comparative study involving 5,585 eyes from 2,793 adult patients with diabetes.
• Handheld retinal images were assessed by POC AI and compared with 5-field RC evaluation.
• Sensitivity and specificity for referable DR (refDR) and vision-threatening DR (vtDR) were calculated.

Main Results:

• Substantial agreement was found between AI and RC for refDR (κ=0.66) and moderate agreement for vtDR (κ=0.54).
• POC AI achieved a sensitivity of 0.86 and specificity of 0.86 for refDR, meeting FDA thresholds.
• POC AI had a sensitivity of 0.92 and specificity of 0.80 for vtDR, not meeting FDA thresholds.
• Image ungradable rates were 7.5% by RC and 15.4% by AI.

Conclusions:

• POC AI demonstrates capability in identifying refDR, meeting FDA standards, potentially improving screening efficiency.
• While not meeting thresholds for vtDR, POC AI integration can reduce RC burden and expedite patient care.
• Implementing POC AI could lead to faster information delivery and more prompt eye care referrals for diabetic patients.