To enhance medical images under low-light conditions, various traditional algorithms such as intensity transformation [21], generative adversarial networks (GANs) [22], Retinex theory [23], histogram equalization [24], and diffusion models [11] have been applied. Fundus image enhancement focuses more on emphasizing high-level structural features relevant to disease diagnosis, such as improving the distinction of lesion areas. Li et al. [25] proposed an algorithm using a global encoder-decoder residual module called NuI-Go to remove uneven illumination in color fundus images. Lee et al. [26] collected a dataset containing high-quality and low-quality fundus images from 2017 to 2019 and proposed a CNN-based fundus image enhancement framework. You et al. [27] proposed an algorithm called CycleCBAM for enhancing color fundus images. This algorithm is an improvement of the CycleGAN network [28], introducing the CBAM module [29] into the CycleGAN network and using cycle consistency loss to constrain the transformation between low-quality and high-quality fundus images. Ma et al. [30] introduced an algorithm named StillGAN, which incorporates a brightness constraint loss to ensure overall brightness balance and optimizes brightness variance to improve the dark regions of fundus images. Hu et al. [31] were the pioneers in proposing fundus image enhancement using diffusion models, utilizing its unique regression mechanism for better enhancement effects. Cheng et al. [32] introduced a diffusion model for enhancing fundus images through degradation learning, finding the degradation mapping from high-quality to low-quality photos using a data-driven degradation framework. Bai et al. [33] presented a network architecture for enhancing low-light medical images using CNNs and diffusion models, leveraging the excellent feature extraction capabilities of CNN and the reverse diffusion process of diffusion models to better enhance low-light medical pictures.