Fractal Sierpinski triangle block division for retina-based glaucoma detection using an optimized hybrid deep...

Area of Science:

• Ophthalmology and Medical Imaging
• Computer Science and Artificial Intelligence

Background:

• Glaucoma is a leading cause of irreversible blindness, often progressing undetected.
• Manual analysis of retinal fundus images is labor-intensive and subject to observer variability.
• Early detection is crucial to prevent vision loss.

Purpose of the Study:

• To develop and validate an automated system for early glaucoma detection using retinal fundus images.
• To combine novel image processing techniques with advanced deep learning models for enhanced diagnostic accuracy.

Main Methods:

• A hybrid deep learning model integrating bidirectional LSTM, GRU, and CNN with attention mechanisms was employed.
• Fractal-inspired Sierpinski triangle spatial decomposition and multi-scale triangular segmentation were utilized.
• Handcrafted features encompassing statistical, frequency-domain, wavelet, morphological, and texture data were extracted.
• Meta-heuristic algorithms, including Harris Hawks Optimizer (HHO), were used for feature subset optimization.

Main Results:

• The HHO-optimized model achieved high accuracy across multiple benchmark datasets (e.g., 98.48% on Drishti-GS, 98.99% on Origa).
• The system demonstrated consistent and reliable performance in identifying glaucoma.
• The automated method proved effective in detecting various glaucoma stages, including early and multi-class scenarios.

Conclusions:

• The proposed automated system offers a scalable, non-invasive, and efficient solution for glaucoma diagnosis.
• This technology supports clinicians in early treatment decisions, potentially improving patient prognosis.
• The fusion of advanced image analysis and deep learning shows significant promise for ophthalmic diagnostics.