Ultrasound imaging is an inexpensive, noninvasive, and radiation-free method frequently used in clinical diagnosis. However, current ultrasound imaging still suffers from a relatively low signal-to-noise ratio and high levels of speckle noise. Deep learning-based ultrasound image enhancement can alleviate this problem. Ultrasound image enhancement improves the quality, clarity, and contrast of images using image processing techniques and algorithms to better demonstrate anatomical structures and lesion information. Although progress has been made, it still faces many challenges, mainly including insufficient processing of detail information and poor noise suppression in traditional methods. This limits the effectiveness of ultrasound images in clinical diagnosis and medical imaging, requiring more effective enhancement techniques to solve. Based on these observations, we propose an image detail feature-preserving Cyclic coherent Generative Adversarial Network (CycleGAN) based on the Universal Ultrasound Foundation Model (USFM) loss network (USCycleGAN). We designed the generator structure to enhance reconstruction capability and preserve the details of ultrasound images, thereby stabilizing CycleGAN training. Additionally, we define a consistency loss for quantifying perceptual changes between images using the USFM loss network. Our approach, tested on well-established datasets, demonstrates superior performance over existing top-tier methods.