This project presents a hybrid deep learning model integrating Depthwise Convolution, Residual Learning, Inception Modules, and Wavelet Analysis to classify retinal diseases from OCT images. The model addresses the limitations of single deep learning architectures by leveraging ensemble methods and optimized training strategies, achieving improved classification accuracy and robustness. Data preprocessing included image resizing, augmentation, and class balancing. The final model was trained and tested on a modified OCT2017 dataset split in a 12:4:1 ratio. Experimental results show that the optimal configuration—50 epochs and a learning rate of 0.001—yielded a validation accuracy of 84.07%. The model was deployed with a GUI using Tkinter, enabling easy real-time prediction of retinal conditions from OCT images.
The DRIW-Net architecture significantly improves the classification of retinal diseases using OCT images, outperforming traditional single-model approaches. Its ensemble framework enhances generalizability and its GUI deployment ensures practical utility for medical diagnostics.
The model's effectiveness may reduce when applied to unseen datasets with vastly different imaging conditions. Additionally, reliance on image-only inputs limits multimodal interpretability. Further validation with clinical datasets is essential.
Future enhancements include expanding to multimodal datasets, incorporating federated learning for privacy-preserving training, integrating with hospital systems for automated diagnosis, and extending to other eye diseases using transfer learning.