Effective detection of polyps at an early stage is crucial for improving the efficacy of computer-aided diagnostic systems. However, low-light endoscopy images often suffer from noise, insufficient visibility, and low contrast, posing significant challenges for image analysis. In this study, we propose a novel image enhancement technique using Generative Adversarial Networks to address these issues. Our method leverages the CycleGAN model, which is trained on diverse natural image datasets and then tested on low-light endoscopy images. Given the lack of ground truth for low-light endoscopy images, we evaluate the quality of the enhanced high-light images using the non-reference image quality metric. Our results demonstrate that the proposed method significantly improves image brightness, contrast, and texture, contributing to more precise polyp detection. Both visual observations and quantitative assessments confirm the superiority of our approach over existing techniques, positioning it as a promising tool for enhancing medical image analysis. Future work will focus on further refining color preservation and integrating end-to-end mapping strategies to enhance the robustness and reliability of the method.