WCFPA-Based Hybrid Model for Age-Related Macular Degeneration Detection in OCT Image

Cataract, glaucoma, and age-related macular degeneration (AMD) are the first three leading causes of blindness worldwide. Among these three, AMD is the leading cause of blindness in the elderly. Automatic screening of AMD is important for both patients and ophthalmologists. To solve this issue, this work presented a hybrid model for AMD detection using OCT images. Initially, the input OCT image is preprocessed by the improved median filter to enhance the image quality by reducing the noise. Segmentation is the subsequent process, done by the active contour method. Subsequently, reflectivity, curvature, thickness, improved texture features, and statistical features are extracted from the segmented image. Particularly, a modified median ternary pattern (MMTP) is proposed for the texture feature extraction. MMTP improves upon conventional MTP by offering better noise resilience and enhanced feature extraction, making it particularly suitable for detecting diseases such as AMD in noisy OCT images. Also, the mean, median, standard deviation, entropy, and variance are the statistical features that are extracted in this work. For detecting the disease, a hybrid classification model is proposed that is trained with the extracted features. The hybrid model is a combination of CNN and bidirectional gated recurrent units (Bi-GRU). Also, to enhance the performance of the hybrid model, the weights of the models are fine-tuned by a water cycle-based flower pollination optimization algorithm (WCFPA). The hybrid CNN-Bi-GRU model is used for classification, leveraging CNN for image feature extraction and Bi-GRU for temporal dependency modeling. The retinal OCT images dataset is used to evaluate the performance of the proposed model, and it was evaluated using accuracy, sensitivity, specificity, and F1-score. The accuracy of the proposed model is high (91%) when compared to the other traditional models such as WCA, FPA, HGSO, AROA, and TOA. Finally, the simulation results show that the suggested approach is more effective than the other conventional approaches.