In recent decades, Diabetic Macular Edema (DME) has emerged as a significant cause of vision loss among diabetic patients due to retinal fluid leakage. To address this challenge, reliable and efficient diagnostic methods are essential. The proposed methodology aims to facilitate early detection through a multi-stage process, including feature extraction, feature selection, and classification.For feature extraction, we introduce the H2A2Net model, which incorporates a Dense Spectral-Spatial Module (DSSM) that employs 3D convolutional DenseNet-inspired layers to extract spectral-spatial features. This is complemented by a Hybrid Resolution Module (HRM) designed to achieve fine spatial detail through a multi-scale process. Additionally, a Double Attention Module (DAM) is implemented to capture global and cross-channel interactions, utilizing both pixel-wise and channel-wise attenuation. Feature selection is conducted using Cuckoo Search Spider Monkey Optimization (CSSMO), which effectively processes both local and global searches to enable efficient selection of high-value features. In the classification phase, a hybrid AdaBoost-Backpropagation Neural Network (BPNN) model is employed, where BPNNs function as weak classifiers whose outputs are iteratively boosted to create a strong ensemble. Experimental results on the CUHK dataset demonstrate that the proposed method achieves an accuracy of 97.4%, a recall of 97.6%, a specificity of 97%, and an F1-score of 98%. These outcomes surpass those of existing state-of-the-art methods, indicating that the proposed approach offers enhanced robustness and efficiency for DME classification.

DME Classification OCT Dataset H2A2 Feature Extraction Feature Selection Accuracy