Autonomous artificial intelligence versus teleophthalmology for diabetic retinopathy

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic Retinopathy (DR) is a leading cause of vision loss in diabetic patients.
• Early detection and treatment of DR are crucial to prevent severe visual impairment.
• Automated screening tools offer potential for wider and more efficient DR detection.

Purpose of the Study:

• To evaluate the diagnostic performance of artificial intelligence (AI) based automated software for DR detection.
• To compare AI software's DR detection accuracy against evaluations by two masked retina specialists.
• To assess the role of AI in screening for more than mild diabetic retinopathy.

Main Methods:

• 201 patients with diabetes mellitus underwent retinography and spectral domain optical coherence tomography (SD-OCT).
• Retinal images were analyzed using two validated AI DR screening software (Eye Art™ and IDx-DR).
• AI diagnoses were compared with assessments by two independent, masked retina specialists.

Main Results:

• Diabetic Retinopathy (more than mild DR) was detected in 18.9% of patients by specialists and 36 patients by AI.
• AI software demonstrated high sensitivity and specificity for detecting more than mild DR.
• AI software had low rates of ungradable images (6.5% for Eye Art, 8% for IDx-DR).

Conclusions:

• AI-based automated software achieves diagnostic sensitivity comparable to expert ophthalmologists for detecting diabetic retinopathy.
• Validated AI software can serve as an effective tool for screening diabetic retinopathy.
• AI shows promise in augmenting clinical diagnosis and improving DR screening efficiency.