Improved machine learning-based glaucoma detection from fundus images using texture features in FAWT and LS-SVM classifier

The second leading cause of blindness worldwide, following cataracts, is Glaucoma. Due to the fact that it is sometimes undetectable in the early stages without routine screening, it poses a serious health risk. Noticeable Glaucoma symptoms sometimes take time to develop. Without medical intervention, the eye condition worsens over time. Identifying and diagnosing Glaucoma requires specialized skills for clinicians. Various factors, such as observer errors and fatigue may influence the clinician's assessment. Our proposed methodology uses a novel method based on Flexible analytical wavelet transform (FAWT), Texture feature, and PCA. First, Fundus images are split into their channels and select green channels. FAWT is used for the decomposition of selected channel images. Second, we extract texture-based features like GLCM and CHIP histogram from green channel images; PCA helps for dimensionality reduction, and then reduced features are ranked by t-value. Finally, high-ranked features are employed in an SVM classifier for Glaucoma recognition. The results of the studies showed that our model performed better than cutting-edge methods for classifying Glaucoma. Using tenfold cross-validation, this model improved its classification accuracy to 96.21%, specificity to 97.16%, and sensitivity to 94.18%. The novel method has proven more effective than current Glaucoma categorization techniques.