Customized DenseNet121 Model for Enhanced Glaucoma Detection: A Deep Learning Approach

Glaucoma, a progressive eye disease that can cause vision loss, requires early detection for effective treatment. This research introduces a novel approach to diagnosing glaucoma using machine learning, specifically a customized DenseNet121 feature extractor. The ORIGA_Gray dataset, consisting of 397 training and 253 test images, was utilized. To address class imbalance and improve model robustness, preprocessing and data augmentation techniques were applied, including grayscale conversion, resizing to 224x224 pixels, and normalization. The model, integrating a customized DenseNet121 with additional layers, enhances feature extraction and mitigates overfitting. It achieved a precision of 99 %, sensitivity of 97 %, specificity of 98%, and an F1-score of 98%, along with an overall classification accuracy of 98%. These results highlight the DenseNet121 framework's effectiveness in distinguishing between healthy and glaucoma-affected images, improving diagnostic accuracy.