Glaucoma diagnosis using Gabor and entropy coded Sine Cosine integration in adaptive partial swarm optimization-based FAWT

Abstract

Purpose

Projections estimate that 112 million people could be influenced by glaucoma by 2040, making it a substantial public health concern and a prominent source of blindness due to optic nerve damage from elevated intraocular pressure. Diagnosis and treatment rely on manual or medical imaging techniques requiring expert supervision. However, early detection through computerized analysis of eye fundus images could help delay total blindness.

Design & Method

This work proposes a modified Flexible Analytical Wavelet Transform based on Adaptive Partial Swarm Optimization for Optimal Parameter Selection (APSO-FAWT). It will help to solve an inequality constraint problem and decompose images into sub-bands. The RGB fundus images are split into three channels at the initial stage. Then, the blue channel is selected for APSO-FAWT-based decomposition because it highlights defects in the retinal nerve fiber layers, aiding glaucoma detection and enhancing nerve fiber visibility. In the second stage, Gabor-based features are extracted from Blue Sub-band images, and the entropy-coded Sine Cosine algorithm is deployed to minimize the dimensions of the extracted features. Then, highlighted features are ranked using the t-value technique, and these features are applied to the LS-SVM to categorize the Glaucoma or Normal images. Additionally, ablation studies were performed to assess the effectiveness of each component within the model.

Outcomes

The model was evaluated using tenfold cross-validation, achieving an Accuracy of 97.42%, Specificity of 98.01%, and Sensitivity of 96.83%. The projected Glaucoma diagnosis model shows improved performance compared to existing methods, offering a promising tool for automated glaucoma detection.