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

Glaucoma is a primary cause of permanent vision loss, and it often gets worse without anybody noticing. This makes it very important to find it early to stop vision loss. Manually evaluating retinal fundus images frequently necessitates considerable effort and is prone to observer-dependent discrepancies. To address these constraints, a new automated method for detecting glaucoma is presented. It combines fractal-inspired Sierpinski triangle spatial decomposition with multi-scale triangular segmentation to reliably identify clinically important areas of the retina. Handcrafted descriptors that include statistical, frequency-domain, wavelet, morphological, and texture-uniformity data are taken from small areas, giving a detailed and unique picture of the structures in the retina. A hybrid deep learning system that combines bidirectional LSTM, bidirectional GRU, and CNN with Bi-LSTM models that have attention layers captures spatial-temporal connections while highlighting essential visual cues for diagnosis. The Harris Hawks Optimizer, Grey Wolf Optimizer, Red Fox Optimizer, and Social Feature Optimizer are all examples of meta-heuristic algorithms that improve the feature subset. The HHO-based version gives the best results. A thorough examination using benchmark datasets shows that the system works very well, with 98.48% accuracy on Drishti-GS, 98.99% on Origa, 97.44% on RimOne-V2, 97.80% on HVD Binary-Class, 99.02% on HVD-Advance, 98.05% on HVD-Early, and 97.11% on HVD Multi-Class trials. The proposed system shows high reliability and stability in glaucoma diagnosis, delivering consistent performance across multiple datasets. It provides a scalable, non-invasive, and efficient solution for automatic detection, supporting doctors in early treatment and improving patients chances of recovery.