An Automated OCT-Based Glaucoma Detection with 3-D ResNet18 and a Convolutional Block Attention Module

In this work, a Deep Learning (DL) technique utilizing a 3D ResNet18 model supplemented with a Convolutional Block Attention Module (CBAM) was proposed to successfully extract volumetric characteristics while prioritizing the most informative spatial areas for classification. The data comprised labeled Optical Coherence Tomography (OCT) volumes processed with normalization, resizing, augmentation, and balanced splitting to minimize data leakage. The experimental results indicated that the model obtained stable performance, with validation and test AUC values of 0.9640 and 0.9495, respectively, and corresponding accuracy rates of 87.88% and 85.59%. Additionally, a comparison with previous models further validated the effectiveness of the current model. These results confirm that the framework can reliably achieve high sensitivity and specificity in separating glaucoma from normal cases.