This study presents an innovative framework for Optical Coherence Tomography (OCT) image analysis, enhancing clinical diagnostics in ophthalmology through a novel integration of deep learning and fuzzy logic. The pipeline addresses speckle noise and low resolution in OCT B-scans using a two-stage preprocessing approach: (1) a cascade of median blur and bilateral filtering for noise reduction, and (2) a custom FuzzyContrastEnhance method that dynamically adjusts contrast in the LAB color space, reducing distortion by 20% compared to traditional methods. The Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) reconstructs high-resolution B-scans (300 \(\times \) 300 pixels) from low-resolution inputs (300 \(\times \) 150/200). A convolutional neural network (CNN), implemented in TensorFlow/Keras, classifies both volume OCT data and individual B-scans into Healthy, Diabetic Macular Edema (DME), or other ocular diseases (e.g., Glaucoma, Macular Degeneration), achieving 99% B-scan accuracy and 92% volume accuracy. Evaluated on a custom dataset from Didavaran Clinic, Isfahan, Iran, an ablation study confirms the synergistic contribution of each stage, with 32% faster execution than baselines. This pipeline aligns with AI-driven medical imaging advancements, offering a robust solution for ophthalmic diagnostics.