Automated diabetic retinopathy detection using optical coherence tomography angiography: a pilot study

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) diagnosis using optical coherence tomography angiography (OCTA) is subjective.
• Developing automated diagnostic tools for DR is crucial for efficient patient management.

Purpose of the Study:

• To design and validate a computer-aided diagnostic (CAD) system for automated diagnosis of non-proliferative diabetic retinopathy (NPDR).
• To assess the accuracy of the CAD system in classifying NPDR using OCTA images.

Main Methods:

• A two-center, cross-sectional study included 106 adults with type II diabetes mellitus.
• A novel CAD system analyzed OCTA macular scans, extracting features like blood vessel density, calibre, and foveal avascular zone (FAZ) size.
• An automated classifier was trained using these features for NPDR diagnosis.

Main Results:

• The CAD system achieved 80.0% accuracy using superficial retinal maps and 91.4% using deep retinal maps.
• Combining both maps yielded an overall accuracy of 94.3%, with a 92.4% area under the curve (AUC).
• High sensitivity (97.9%) and specificity (87.0%) were reported, demonstrating robust diagnostic performance.

Conclusions:

• Automated NPDR diagnosis using OCTA images is feasible and highly accurate.
• The developed CAD system offers a reliable tool for objective DR assessment.
• Future integration with OCT data may further enhance the system's robustness.