Retinal diseases pose a significant global public health issue, impacting individuals on personal, familial, and global scales. This study aims to develop a deep learning-based diagnostic tool to classify retinal conditions—Choroidal Neo Vascularization, Diabetic Macular Edema, Drusen, and Normal—using Optical Coherence Tomography or OCT scans. A key component is the preprocessing technique, which enhances image quality, reduces noise, detects edges, and applies thresholding and gradient-based contour detection. The proposed architecture employs CNN models with three, four, and five layers, achieving accuracy rates of 53.50%, 78.50%, and 90.00%, respectively, with the five-layer model performing best. Additionally, preprocessed images were tested with pre-trained models VGG16 and ShuffleNet, achieving accuracy rates of 96% and 76.12%, respectively. By combining advanced preprocessing and neural network training, we accurately categorized retinal disorders, showcasing the potential of this approach for improving diagnostic accuracy in retinal disease detection, thereby offering a promising tool for early diagnosis and better patient outcomes.