This research presents a hybrid deep learning framework combining CycleGAN and CNN to improve both the quality and classification of ultrasound images. The CycleGAN model is first applied to enhance low-resolution ultrasound scans by removing artifacts while preserving anatomical structures. The refined images are then classified into benign, malignant, and normal categories using a Convolutional Neural Network. Quantitative evaluation using PSNR and LNCC confirms significant improvements in image clarity and structural fidelity. The CNN trained on enhanced images achieved high precision and recall, outperforming conventional methods. This study demonstrates the effectiveness of integrating GAN-based enhancement with deep classification for advancing diagnostic accuracy in medical imaging.