Glaucoma Diagnosis via Advanced Retinal Image Processing and V-RACNet Classification

The proposed paper presents a general architecture of processing context in glaucoma diagnosis, entailing steps such as image preprocessing to the classification with the new V-RACNet model. The initial stages involve image resizing, removal of the green plane, and enhancement of blood vessels extraction by morphological operations and thinning. Focal regions and features extraction of the fovea regions including GLCM texture features and statistical properties are identified through a clustering algorithm (K-means). The glaucoma detector trained on samples of 1114 samples of DRIVE database with 650 normal samples and 464 glaucomatous samples (Low-Tension Glaucoma or Angle-Closure Glaucoma). The dataset is separated into two portions, training (70%, 779 samples) and testing (30%, 335 samples) to determine the model performance and ability to generalize its performance on new data. Further on, a V-RACNet model is constructed, and trained on a labelled database, and tested to classify glaucoma. The proposed method has remarkably high accuracy (99.32%), sensitivity (99.42%), specificity (98.58%), and precision (99.26%) which indicates that it can be a powerful tool because of its automatization capabilities in glaucoma diagnosis.