Knee osteoarthritis (KOA) is a progressive joint disorder often diagnosed via X-ray imaging due to its accessibility and low cost. The severity is typically assessed using the Kellgren and Lawrence (KL) grading scale. Early detection using this method can slow disease progression through timely intervention. In this study, four datasets were combined and augmented using Deep Convolutional Generative Adversarial Networks (DCGAN), resulting in 110,232 enhanced images. Advanced preprocessing techniques such as adaptive histogram equalization (AHE), fast non- local means filtering, and resizing were applied to improve image quality. A modified Compact Convolutional Transformer model, KOA-CCTNet, was developed and fine-tuned by optimizing multiple hyperparameters to efficiently handle large-scale data and reduce training time. Experimental results showed KOA-CCTNet outperformed state-of-the-art models including ResNet50 (80.77%), MobileNetv2 (79.98%), DenseNet201 (80.23