Capsule network-based deep learning for early and accurate diabetic retinopathy detection

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Glaucoma is an optic nerve disease leading to blindness if untreated, posing diagnostic challenges.
• Current diagnostic methods rely on analyzing fundus images for optic disc and cup features.
• Artificial intelligence (AI) offers potential for enhancing glaucoma identification accuracy.

Purpose of the Study:

• To implement a hybrid AI model using UNet++ and Capsule Network (CapsNet) for improved glaucoma diagnosis.
• To leverage UNet++ for accurate semantic segmentation of optic discs (ODs) and optic cups (OCs).
• To utilize CapsNet for capturing hierarchical structures and enhancing sensitivity to glaucomatous changes.

Main Methods:

• Retinal images were pre-processed using Histogram Equalization and Contrast Limited Adaptive Histogram Equalization (CLAHE).
• A hybrid model combining UNet++ for segmentation and CapsNet for feature analysis was developed.
• The model was trained and tested on ten benchmark datasets.

Main Results:

• The hybrid model achieved high accuracy in optic disc and cup segmentation.
• The system demonstrated superior performance in glaucoma detection compared to existing approaches.
• Evaluation indicated good diagnostic ability, paving the way for automated glaucoma diagnosis.

Conclusions:

• The UNet++ and CapsNet hybrid model offers a novel and efficient method for glaucoma diagnosis.
• Early detection of glaucoma through this AI application has the potential to prevent blindness.
• AI is revolutionizing ophthalmic healthcare, particularly in diagnosing conditions like glaucoma.