Screening Glaucoma With Red-free Fundus Photography Using Deep Learning Classifier and Polar Transformation

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is a leading cause of irreversible blindness worldwide.
• Early detection of glaucomatous retinal nerve fiber layer (RNFL) defects is crucial for timely intervention and vision preservation.
• Current diagnostic methods may have limitations in sensitivity and accessibility.

Purpose of the Study:

• To develop novel software for automated detection and localization of RNFL defects in fundus images.
• To utilize a deep learning classifier and polar transformation technique for enhanced diagnostic accuracy.

Main Methods:

• A deep learning classifier was employed to extract bottleneck features and output glaucoma probability.
• An image processing algorithm involving contrast enhancement, region of interest selection, polar image conversion, and blood vessel removal was implemented for RNFL defect localization.
• The algorithm calculated average curvatures and identified defects based on a defined cut-off value.

Main Results:

• The study included 100 normal controls and 100 open-angle glaucoma patients.
• A significant difference in maximum curvature was observed between control and glaucoma groups (P<0.001).
• The deep learning classifier achieved an area under the receiver operating characteristic curve of 0.939, outperforming the maximum curvature method (0.711).

Conclusions:

• The proposed software demonstrates effectiveness in the automated detection of RNFL defects.
• The integration of deep learning and polar transformation offers a promising approach for economical and effective early glaucoma diagnosis.